Molecular Imaging through Magnetic Resonance for Clinical Oncology

Karen Belki

CISP

CAMBRIDGE INTERNATIONAL SCIENCE PUBLISHING

Molecular Imaging through Magnetic Resonance for Clinical Oncology

Karen Belki
Department of Oncology - Pathology Karolinska Institute, Sweden and Institute for Health Promotion & Disease Prevention Research, University of Southern California, U.S.A.

Cambridge International Science Publishing Cambridge, Great Britain

Copyright 2004, by Karen Belki All Rights Reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic, mechanical, recording, or otherwise, without the prior written permission of the authors.

CAMBRIDGE INTERNATIONAL SCIENCE PUBLISHING Cambridge CB1 6AZ Great Britain http://cisp-publishing.com A catalogue record for this book is available from the British Library ISBN: 1-904602-29-0

About the author

Dr. Karen (Edinger) Belkic was born in Los Angeles, California. She is an Adjunct Associate Professor of Preventive Medicine at the University of Southern California School of MedicineInstitute for Health Promotion and Disease Prevention Research, and currently is a visiting Clinical Researcher and Pedagogue at the Department of Oncology-Pathology, Karolinska Institute, Stockholm, Sweden. She received her B.A. in biology from the University of California, Santa Barbara and her M.D. degree from the University of Southern California School of Medicine. She holds a PhD in the neurosciences and is a physician specialist in Internal Medicine. Dr. Belkic has served as a physician within the fields of occupational medicine, internal medicine and cardiology. Dr. Belkic has authored over 60, widely cited peer-reviewed articles and book chapters, one book and four monographs in multi-disciplinary areas related to preventive and internal medicine. She is a member of the Editorial Board (Responsible for Computational Medicine) for the Journal of Computational Methods in Sciences & Engineering (JCMSE), Cambridge International Science Publishing. She has twice been an invited guest co-editor of special issues in international journals, most recently on How Mathematical Advances can Optimize Magnetic Resonance Spectroscopy in Oncology for the JCMSE. A major thread tying together the clinical research activity of Dr. Belkic has been the search for non-invasive, sensitive and specific tools for early detection and prevention of disease. Her current clinical research activity is focused upon incorporating advances in signal processing for biomedical imaging to take fuller advantage of the possibilities offered by MRSI for surveillance imaging in oncology, to optimize diagnostics and fine-tune therapy within a multi-level patient-centered strategy in which quality of life, treatment and clinical monitoring are part of an integrated whole. She has been exploring how these advances might be implemented as part of screening surveillance for early cancer detection, particularly for malignancies that afflict women. Dr. Belkic has taken a broad view, looking not just at the immediate (i.e. proximal) markers of risk, but taking into account the more distal, and potentially key, determinants of disease. Thus, she has been very interested in how the environment (especially the work environment) impacts upon target organs, often mediated by the central nervous system. Within this framework, Dr. Belkic has developed multi-level models. These incorporate, inter alia, non-linear, parametric methods in signal processing in relation to multiple physiological time signals for functional diagnostic testing. Dr. Belkic has a special interest in pedagogy. She has focused much of her current activity on how to incorporate a multi-disciplinary approach into the medical school curriculum, whereby a firm grounding in clinical medicine is integrated with an appreciation of biology, physics and mathematics, and to place this within the larger framework of comprehensive care and prevention.

CONTENTS
Preface Acknowledgments i vii

Introduction: Molecular Imaging and Oncology

1.1 1.2 1.3 Functional and anatomical imaging: The importance of combined modalities for oncology Spectroscopic imaging through magnetic resonance: Possibilities to be fully tapped for clinical oncology Objectives and organization of this book 1 2 3

Part A: A Brief Overview of Basic Principles

2 Magnetic Resonance
2.1 2.2 Precession: Response of proton hydrogen and some other nuclei to a static magnetic field Response to a 90 radio-frequency pulse: Resonance and relaxation 2.2.1 Free induction decay 2.2.2 T1 and T2 relaxation and weighting 2.2.3 Spin-echo versus gradient echo sequences Technical considerations 2.3.1 Components of a clinical MR scanner 2.3.2 Motion artefacts 8 11 13 14 17 17 18

2.3

Magnetic Resonance Spectroscopy (MRS)

3.1 Chemical shift as the basis for identifying metabolites 21

3.2
3.3.

Characteristics of the metabolites evaluated by MRS

1 H (proton) MRS 3.3.1 The in vivo proton MR spectrum 3.3.2 Suppression of giant resonances 31 P MRS MRS of a single volume 3.5.1 Voxel positioning 3.5.2 Single-voxel sampling methods Magnetic field inhomogeneity 3.6.1 Relation of spin-echo to B0 inhomogeneity 3.6.2 Shimming

24
25 26 27 28 29 30 30

3.4 3.5

3.6

Basics of Signal Processing for Magnetic Resonance

4.1 4.2

4.3

4.4

The concept of conjugate variablesComplementary representations From the time to the frequency domain 4.2.1 Familiar clinical examples outside MRS 4.2.2 Spectral analysis of a decaying time signal (FID) Spatial localization: from the momentum to the coordinate representations 4.3.1 Gradient magnetic fields: From k space to MR image A brief mathematical background 4.4.1 The Fourier transform 4.4.2 How a gradient field yields spatial localization 4.4.3 Recent advances in signal processing relevant to MR: The fast Pad transform

33 34 35

39 42 45 47

Magnetic Resonance Spectroscopic Imaging (MRSI)

5.1 5.2 Spatial resolution of spectra Implementation of MRSI 5.2.1 Reconstruction and post-processing of MRSI data 5.2.2 An illustration of MRSI 57 59 60 60

Safety Considerations
6.1 6.2 6.3 Effects of exposure to electromagnetic fields & other MR risks 63 Precautions in MR 65 Contraindications to MR 65

Part B: MRS and MRSI--Current State of the Art in Clinical Oncology

7 Magnetic Resonance in Cancer Diagnostics Generalities
7.1 7.2 7.3 Relaxation rates of neoplastic tissues Contrast enhancement Informative metabolites and ratios from in vivo MRS 7.3.1 1H MRS 7.3.2 31P MRS MRSIImportance of full volumetric coverage 69 70 71 72 72

7.4

Brain Tumor Diagnostics

8.1 8.2 Overview of epidemiological and clinical aspects MRI in brain tumor diagnostics 8.2.1 Characteristics on T2 weighted imaging 8.2.2 Characteristics with contrast enhancement 8.2.3 Diffusion weighted imaging Primary diagnosis of brain tumors with 1H MRS/MRSI 8.3.1 Comparisons with normal brain tissue 8.3.2 Comparisons with non-neoplastic brain lesions MRS and MRSI in grading of brain tumors MRS and MRSI for histopathologic classification MRSI for brain tumor localization 75 82 83 83 85 94 98 102 108

8.3

8.4. 8.5. 8.6

8.7 8.8 8.9

Gauging response to treatment with MRS/ MRSI Prognostic information provided by MRSI 2D J-resolved MRS and in vitro MRS: Further insight into metabolic characteristics of brain tumors

109 110 111

Prostate Cancer Diagnostics

9.1 9.2 Overview of epidemiological and clinical aspects MR-assisted prostate cancer diagnostics 9.2.1 MRI for diagnosis and staging 9.2.2 MRSI for primary diagnosis MRI and MRSI for treatment planning Follow-up with MRI and MRSI after treatment 2D COSY and in vitro MRS studies of prostate cancer 121 129 130 136 139 141

9.3 9.4 9.5

10

Gynecologic Cancers
10.1 Ovarian Cancer 10.1.1 Overview of epidemiological and clinical aspects 149 10.1.2 Current approach to 1 diagnosis and staging 154 10.1.3 In vivo MRS in ovarian cancer 158 10.1.4 In vitro MRS findings in ovarian cancer 160 Cancer of the Uterine Cervix 10.2.1 Overview of epidemiological and clinical aspects 163 10.2.2 Current approach to 1 diagnosis and staging 165 10.2.3 In vivo MRS in cervical cancer 166 10.2.4 In vitro MRS findings in cervical cancer 168 Endometrial Cancer 10.3.1 Overview of epidemiological and clinical aspects 169 10.3.2 Current approach to 1 diagnosis and staging 172 10.3.3 In vitro MRS findings in endometrial cancer 175 Comment on differences in approach to gender-specific cancers using MRS and MRSI 176

10.2

10.3

10.4

11

Head and Neck Cancer

11.1 11.2 Overview of epidemiological and clinical aspects Anatomical and Functional Imaging 11.2.1 FDG-PET and CT 11.2.2 MRI 11.2.3 In vivo MRS for 1 detection and assessing response to therapy 11.2.4 In vitro MRS studies 181 189 189 191 192

12

Non-Hodgkins Lymphoma (NHL)

12.1 12.2 Overview of epidemiological and clinical aspects Anatomical and Functional Imaging 12.2.1 FDG-PET and CT 12.2.2 MRI 12.2.3 MRS assessment of lymph node involvement and response to therapy 31 P-MRS assessment of hepatic lymphoma 12.2.4 31 P-MRS assessment of testicular lymphoma 12.2.5 197 208 209 209 213 213

13

SarcomasMusculoskeletal tumors
13.1 13.2 13.3 Overview of epidemiological and clinical aspects Current approach to 1 diagnosis and staging MRS in assessment of sarcomas 13.3.1 Primary detection 13.3.2 Assessment of response to therapy 13.3.3 In vitro MRS studies of sarcomas 217 223 224 225 227

14

Renal Cell Carcinomas (RCC)

14.1 14.2 14.3 Overview of epidemiological and clinical aspects Current approach to 1 diagnosis and staging Initial studies using in vivo 1H MRS in RCC In vitro MRS studies of RCC 233 236 237 237

14.4

15

Hepatic, GI and other Tumors

15.1 Functional anatomical imaging using PET and advanced CT-based methods 15.1.1 FDG-PET for cancerous involvement of the liver 15.1.2 FDG-PET & virtual CT endoscopy in GI cancers 15.1.3 FDG-PET for diagnosis of some other cancers MRI and MRS in hepatic cancers 15.2.1 Hepatocellular carcinoma 15.2.2 Hepatic lymphomas 15.2.3 Liver Metastases 15.2.4 MRS studies on heterogeneous liver cancers Magnetic resonance diagnostics in colorectal cancer 15.3.1 Virtual MR colonography 15.3.2 In vitro MRS findings Magnetic resonance diagnostics in gastric cancer 15.4.1 Virtual MR gastroscopy 15.4.2 In vitro MRS findings Magnetic resonance diagnostics in esophageal cancer 15.5.1 MRI and MR endoscopy of the esophagus 15.5.2 In vitro MRS findings Magnetic resonance diagnostics in pancreatic cancer 15.6.1 MRI and MR endoscopy of the pancreas 15.6.2 Animal MRS studies of pancreatic cancer In Vitro MRS applied to some other cancers 15.7.1 Malignant melanoma 15.7.2 Cancer of the thyroid 243 244 245 247 248 249 249 250 252 253 255 256 256 257 258 258 259 259

15.2

15.3

15.4

15.5

15.6

15.7

16

Breast Cancer Screening and Early Diagnosis

16.1 16.2 16.3 16.4 Overview of epidemiological and clinical aspects Breast cancer screening with mammography Molecular imaging with FDG-PET and scintimammography Magnetic resonance: Early results in 1 breast cancer diagnosis 16.4.1 MRI 1 H MRS 16.4.2 16.4.3 In vitro 1H MRS findings: Detailed comparison of metabolite concentrations in breast cancer vs. nonmalignant breast tissue 265 271 273 274 275 275

Part C: Future Perspectives for MRS and MRSI in Cancer Diagnostics

17 Limitations of MRS and MRSI in Oncology: Relation to Reliance on the Conventional Framework for Data Analysis
17.1 17.2 17.3 17.4 17.5 Low Resolution of the Fast Fourier Transform (FFT) Poor signal-to-noise ratio (SNR) of the FFT The FFT supplies only a shape spectrum The FFT requires fitting number of metabolites guessed in advance A small number of observable compounds with FFT 286 287 289 291 293

18

Mathematical Advances in Spectral Analysis: Potential Relevance for Cancer Diagnostics using MRS and MRSI
18.1 18.2 18.3 18.4 18.5 18.6 18.7 18.8 18.9 The fast Pad transform (FPT) has rapid, stable convergence 299 FPT improves resolution and SNR 300 FPT determines the exact number of metabolites 302 FPT treats Lorentzians and non-Lorentzians 303 FPT unambiguously identifies overlapping resonances 303 FPT accurately estimates concentrations 303 An illustration of the performance of the FPT for a clinical MRS signal 304 Validity of the FPTerror analysis 306 Appropriateness of the FPT for tumor diagnostics 306 within magnetic resonance

19

Next needed Steps--Application of the FPT with the Aim of Optimisation of MRSI in Cancer Diagnosis
19.1 View to Early Detection and Screening

19.1.1 19.1.2

Challenges in cancer screening Optimization of MRSI for early detection of cancers

310 311

20

Concluding Comments and Outlooks: Prevention, Early Detection, and Monitoring of Cancer in a More Comprehensive Perspective
20.1 Emphasis upon quality of life for patients with cancer 20.1.1 Importance for return to healthy work 317 318

20.1.2

A patient-centered comprehensive strategy

319 321 327

Appendix 1 Glossary of Terms Relevant to MRS and MRSI Appendix 2 List of Acronyms

Preface

Molecular imaging represents a new medical discipline that aims at uncovering molecular pathways of disease by integrating cellular and molecular biology with diagnostic imaging1,2. It encompasses several of the exciting imaging modalities that are capable of providing critical information for early detection and progression of disease, through its cellular and molecular pathways. Given the rapidly growing sophistication in our understanding of the cell biology of cancer, molecular imaging offers a strategic bridge to clinical oncology. It is of critical importance to non-invasively assess the behavior of tumor cells in their milieu, for which in vivo Moreover, in vivo molecular imaging is a key methodology3. molecular imaging detects not only the presence of the disease process, but can also quantify its extent and severity, as well as following the course of disease over time4. Thus, with respect to oncology, in addition to its proven diagnostic potential, molecular imaging is also vital to target definition for dose delivery in e.g. radiotherapy. Molecular imaging is also an invaluable tool used during and after therapy for assessing dose delivery and the overall success of treatment. In the very recent period, molecular imaging has undergone fundamental change, from being almost synonymous with clinical positron imaging, to embrace a much broader scope, which now also encompasses magnetic resonance spectroscopy and imaging At this and the exquisite information offered thereby1. crossroads, it becomes an urgent task for clinical oncology to
J.R. Barrio, Editorial, Molec. Imag. Biol. 4, 267-273 (2002). F.D. Rollo, Molecular imaging: an overview and clinical applications, Radiol. Manage. 25, 28-32 (2003). 3 D.C. Sullivan, Challenges and opportunities for in vivo imaging in oncology, Technol. Cancer Res. Treat. 1, 419-422 (2002). 4 M. Schwaiger, F.M. Bengel, From thallium scan to molecular imaging, Molec. Imag. Biol. 4, 387398 (2003).
1 2

ii

determine the optimal potential, role and complementarity of the various molecular imaging methods, namely those from nuclear medicine as compared to those based upon magnetic resonance, and upon ultrasound. This question is addressed in the present book, focusing upon magnetic resonance spectroscopy (MRS) and spectroscopic imaging (MRSI), with its numerous advantages for oncology -- superior contrast among tissues, multiplanar capabilities, together with the potentially rich array of spectral characteristics that distinguish malignant from healthy tissues, and lack of exposure to ionizing radiation. The combination of anatomic localization and insight into metabolic characteristics from spectral information is often decisive for accurate and timely identification of malignancy. This has proven invaluable, especially in the most difficult cases, e.g. differentiating recurrent tumor from radiation necrosis or post-operative changes. These advantages have become particularly clear in neuro-oncology, where MRS and MRSI are now a key modality for nearly all aspects of brain tumor diagnostics. In fact, in no other area of oncology have MRS and MRSI become so widely incorporated into clinical practice. There is a literal explosion of information on MRS and MRSI in neuro-oncology in the last 2 years. This book provides a state-of-the-art review of this body of knowledge, which, indeed, represents an important advance for the detection and characterization of tumors of the brain. To highlight a few of the most striking achievements: MRS-guidance has improved the yield of diagnostic tissue obtained with stereotactic brain biopsy; metabolic maps from MRSI have helped refine radiation therapy dose contouring, as well as surgical planning to treat brain tumors more effectively. With MRS and MRSI, brain tumor recurrence is now detected with greater sensitivity than was previously the case with magnetic resonance imaging (MRI) alone. MRS and MRSI have also made great strides in prostate cancer diagnostics, as systematically appraised in this book. Compared to MRI alone, MRSI substantially improves the diagnostic accuracy for detection of prostatic tumor and extracapsular extension, as well as helping to distinguish cancerous prostate from benign glandular hypertrophy. Other areas of clinical decision-making in prostate cancer have been impacted as well, including treatment planning with brachytherapy, and assessing tumor regression versus recurrence after treatment. Some useful initial results with MRS have been obtained for identifying and characterizing certain gynecological tumors (ovarian and uterine cervix). However, the difference in approach to genderspecific cancers using MRS and MRSI is perhaps the most striking observation emerging from the review of this topic. Namely, one is struck by the contrast between the wealth of studies using in vivo

iii

MRS and MRSI for prostate cancer with noteworthy improvements in diagnostic accuracy compared to the relative dearth of publications applying in vivo MRS and MRSI to gynecological cancers. Head and neck cancers are a heterogeneous group of tumors, which have heretofore posed substantial challenges in interpretation of histopathologic findings. Molecular imaging through MRS has proven to be of benefit in the primary diagnosis of these diverse cancers, as well as for monitoring the effectiveness of therapy. It may be particularly helpful in identifying post-therapeutic changes (post-operative fibrosis and radiation necrosis), which have heretofore been extremely difficult to distinguish from recurrent disease. Lymphomas are also a diverse group of malignancies, whose incidence in many parts of the world is rising faster than virtually any other cancer. From this book is it seen that molecular imaging via positron-emission tomography (PET) and via MRS has been of substantial clinical importance for lymphomas in assessing disease extent and response to therapy. Notably, both of these molecular imaging modalities are found to be superior to their anatomical counterpart methods, i.e. computerized tomography and MRI, in the timely identification of responders versus non-responders to therapy. This is particularly critical for clinical decision-making, such as when to begin induction chemotherapy or to change treatment regimen. In other words, molecular responses frequently preceded morphologic changes, thereby providing earlier insight into the effectiveness of therapy. While sarcomas are quite rare in the general population, certain groups, including those who have been treated either by radiation or some chemotherapeutic agents, are at increased risk for developing these tumors. Thus, this becomes an important area for follow-up surveillance in oncology, especially for survivors of childhood cancers. MRS and MRSI have shown some utility in the primary detection of sarcomas. MRS has also provided substantive insights concerning response to therapy in advanced stages of these cancers, where decisions are fraught with the greatest difficulty, e.g. whether or not to proceed with isolated limb perfusion and limbsparing resection. The approach to primary diagnosis and staging of renal cell carcinomas has undergone dramatic improvement, with the rapid developments in CT and MRI technology. Preliminary results with molecular imaging through MRS are promising here, as well. Molecular imaging is in the forefront of a number of other abdominal cancers. Liver metastases are now most accurately

iv

diagnosed through PET imaging, while contrast MRI is currently the method of choice for hepatocellular carcinoma. In vivo MRS and MRSI are helpful not only in distinguishing hepatic malignancies from normal liver, but also in assessing response to chemoembolization. While conventional endoscopy continues to be the gold standard for primary diagnosis of cancers of the gastrointestinal tract, there are exciting developments in virtual endoscopy through MR as well as CT imaging. These advanced MRI techniques are reviewed, as well, in this book. Of particular note are the possibilities for these virtual imaging techniques to detect lesions that are difficult to reach and therefore sometimes missed by conventional endoscopy. While in vivo studies of the gastrointestinal tumors are sparse, some in vitro investigations have been carried out, e.g. on esophageal cancer. These hold some promise to help predict the biologic behavior of these tumors. The result of this line of investigation may be improved selection of patients who would benefit from specific treatment approaches such as neoadjuvant and post-operative therapies. Worldwide, breast cancer is the most commonly occurring malignancy in women, and the leading cause of cancer-related deaths among women. Regular screening with mammography reduces breast cancer mortality, but has serious limitations in diagnostic accuracy, plus entailing exposure to ionizing radiation. As reviewed in this book, a number of molecular imaging modalities are being actively explored with respect to various aspects of breast cancer diagnostics. MRI and MRS are being examined as adjuncts, or possibly even alternatives to mammography, especially for younger women at high risk. Molecular imaging through MRS has been shown to improve the specificity of MRI for the diagnosis of breast cancer, and is attracting attention as a promising new technique for improving primary detection of breast cancer. Having just very briefly outlined some highlights from the middle portion of this book, the impression could certainly be drawn that molecular imaging through MRS and MRSI has achieved important strides in cancer diagnostics. And, indeed, there is no question whatsoever that this is the case. But that is only part of the story. The question is: what could MRS and MRSI offer to clinical cancer diagnostics? In other words, have these modalities realized their potential in the battle against the scourge of cancer? It is here, that this book differs fundamentally from many other texts. Not only is the answer a resounding no, but the focus is not upon upcoming technological, i.e. hardware, advances, important as these may be. Rather, the reader including the clinician, clinical researcher and medical student, is invited to join in the exploration of what mathematical advances in signal processing could offer for molecular imaging through MRS and MRSI.

The motivation for this approach is that, notwithstanding the achievements, there are still important shortcomings of the present applications of in vivo MRS and MRSI in oncology. These are explored in detail. First of all, very few of the currently assessed metabolite concentrations or ratios unequivocally distinguish tumors from normal tissue. The spectroscopic criteria, e.g. cut points used to define brain or prostate cancer, vary widely from author to author. Moreover, changes in each of the metabolite concentrations and ratios are non-specific for cancer. For example, non-neoplastic processes such as infection, stroke, demyelinating disorders, inter alia, frequently show spectral changes that are identical to those seen in brain tumors. Histopathological characterization and brain tumor grading, both crucial for clinical decision-making, have been greatly aided by MRS and MRSI. However, there are numerous contradictory findings in the literature. Particularly troublesome is the limited possibility of MRS and MRSI to detect very small tumors, at the very time at which therapeutic interventions would have the best chance for success. In this book it is clearly demonstrated that many of the inadequacies of MRS and MRSI for clinical oncology are directly related to the reliance upon the conventional data analytical method, i.e. the Fast Fourier Transform which has low resolution, poor signal-to-noise ratio (SNR) in clinical signals, supplies only shape spectra and requires fitting, which is non-unique, so that the number of metabolites must be guessed in advance. This can lead to spurious peaks (over-fitting) and true metabolites being undetected (under-fitting). These limitations can be circumvented by recent mathematical advances in signal processing via e.g. the Fast Pad transform (FPT)5. As a high resolution, non-linear, stable parametric method, the FPT substantially improves SNR, and fulfills stringent requirements for tumor diagnostics: no post-processing fitting, provides precise numerical results for all peak parameters, and specifies the exact number of metabolites (including those that overlap) from the encoded data with rapid and stable convergence. As reviewed in detail, in vitro MRS findings for a wide array of human cancers clearly distinguish malignant and normal tissues, and frequently offer insights into molecular mechanisms. The most important information for detecting malignant lesions is often found in closely overlapping resonances, some of which decay rapidly and therefore can only be detected at short acquisition times.
5

Dz. Belkic, Quantum Mechanical Signal Processing and Spectral Analysis, Institute of Physics Publishers, Bristol, 2004.

vi

A clinical MRS signal is presented which illustrates the advantages of the fast Pad transform, in providing rapid and stable convergence with accurate data about key metabolites and their concentrations at short acquisition times. Thereby, it is shown that the FPT offers greater possibilities to extract the rich spectroscopic information, which heretofore has remained untapped with the conventional Fourier approach. This book has been designed to be self-contained, and does not presume prior knowledge about spectroscopy or imaging. The basic principles of MRI and then MRS and MRSI are explained in the first chapters. Relevant computational aspects are included. Signal processing is initially presented intuitively, using a conceptual approach to conjugate variables, with familiar clinical examples. This is followed by a succinct mathematical exposition aimed at a broad audience. The reader will thereby gain an appreciation of the strategies that allow MR signals to be transformed into clinically meaningful information. A brief overview of key technological aspects of MR and the nomenclature is also given. A brief background is also provided about the epidemiological and clinical aspects of the tumor types examined with MRS and MRSI. Other diagnostic modalities and typical findings using MRI are presented, as well, so that the reader will have a comprehensive view of how these tumors are currently identified. Within this background, emphasis is placed upon etiology and risk factors, in order to suggest when the clinical index of suspicion should be raised, and when, therefore, the information provided by MRS and MRSI might be of particular benefit. In the concluding chapters of the book, some possibilities are projected via optimized MRS and MRSI for improving surveillance of patients with or at risk for specific cancers. The rationale for this approach is to help achieve the positive aims of screening. It is suggested that there is a need to consider advances in cancer diagnostics as provided by molecular imaging within a broader perspective. Overall, this can be seen as a multi-level patient-centered global and comprehensive strategy in which quality of life, treatment and clinical monitoring are part of an integrated whole. Karen (Edinger) Belki Adjunct Associate Professor of Preventive Medicine Physician Specialist in Internal Medicine Karolinska Institute and University of Southern California School of Medicine Stockholm, Sweden October 2004

vii

Acknowledgments

The author would like to thank Professor Devad Belki for critical review of this book and fruitful discussions. Thanks are due to Victor Riecansky from Cambridge International Science Publishing (CISP) for his cooperation. The work on this book was supported by the following four Swedish funds: the Stockholms County Council through FoUU at the Karolinska Hospital, the Swedish Scientific Research Council (Vetenskapsrdet), the Swedish Cancer Society (Cancerfreningen) and King Gustav the Fifths Jubilee Foundation.

This book is devoted to our patients and their families, in admiration of their courageous struggle.

Chapter 1

Introduction: Molecular Imaging and Oncology

_______________________________________________________________________________

Molecular imaging is impacting dramatically upon virtually all areas of clinical medicine. As stated by Tepper, we now appear to be heading to a watershed where biology and imaging come together, to let us use information from both areas in clinical decision making and treatment implementation. The advances in functional imaging have begun to produce, and will continue to produce, major changes in our day-to-day practice of medicine [1] (p. 547). This is particularly true for oncology. The European Journal of Cancer recently devoted a special issue to the explosion of new information created by molecular imaging [2]. The recent advances in molecular imaging are viewed by the National Cancer Institute (NCI) of the USA as an extraordinary opportunity for early detection by identifying the key changes for the emergence and progression of cancer on the molecular and/or cellular level [3].

1.1 Functional plus anatomic imaging: The importance of combined modalities for oncology
Functional imaging is of greatest value for oncology when combined with anatomic information. Currently in clinical oncology, the most common combination is positron emission tomography (PET) plus Computerized Tomography (CT) yielding PET-CT. Kluetz and colleagues note [4]: Simultaneous acquisition of co-registered PET and CT images enables physicians to more precisely discriminate

between physiologic and malignant FDG1 uptake and more accurately localize lesions, improving the value of diagnostic PET in oncologic applications (p. 223). The impact of FDG-PET and PET-CT upon detection, staging and management of patients with cancer has been profound (see e.g. reviews [5, 6]), such that this diagnostic tool has become part of the routine armamentarium for the assessment of many patients with proven or suspected cancer [6] (p. 35).

1.2 Spectroscopic imaging through MR: Possibilities to be fully tapped for clinical oncology
Functional-anatomic imaging can be achieved as well, by combining Magnetic Resonance Imaging (MRI) and Magnetic Resonance Spectroscopy (MRS) yielding MRSI (Magnetic Resonance Spectroscopic Imaging). The images produced by MRI alone yield anatomical information depicted by the spatial distributions of magnetizations of nuclei from the scanned tissue as a result of the effects of three external magnetic fields, as will be detailed in the first part of this book. One of the key advantages of MRI is superior contrast among tissues compared to CT. Moreover, MRI has multiplanar capabilities, that are absent from CT [7]. Via magnetic resonance, information can also be obtained about many of the chemical constituents of biological material. This is achieved through MRS, which involves the in vivo application of traditional laboratory-based NMR2 techniques. MRS provides this complementary biochemical and physiologic information in the form of spectra. As yet, however, application of MRS and MRSI in tumor diagnostics, although increasing, has been substantially more limited than PET and PET-CT. This might seem surprising, given that in vitro studies often describe a rich array of spectral characteristics that distinguish malignant from healthy tissues. Therefore, MRSI should be able to identify key biochemical changes, much before the tumor becomes detectable by other functional imaging methods that mainly rely upon single markers that are not entirely sensitive or specific for malignant activity.

1 FDG = 18F-deoxy glucose is a glucose analogue which is phosphorylated by hexokinase (the first enzyme of the glycolytic pathway) to FDG-6-PO4 and remains trapped intra-cellularly, thus reflecting glycolysis throughout the body. Since many tumor cells have a high metabolic rate mainly via the glycolytic pathway, FDG uptake has been used as a marker of malignancy. 2

NMR = nuclear magnetic resonance

Molecular imaging through magnetic resonance could be particularly well suited for screening and repeated monitoring since it entails no exposure to ionizing radiation. Perhaps part of the reason why, with a few important exceptions, MRS and MRSI are not yet part of the routine armamentarium for the assessment of patients with proven or suspected cancer is historical. Namely, early in vivo spectrometers did not have imaging capabilities, such that MRS and MRI developed along separate paths [8]. Furthermore, as pointed out over a decade ago by Bottomley [9], spectroscopy differs fundamentally from its sister technology, MR imaging, to which it is inextricably linked, {such} that its failure to materialize clinically with the same speed that MR imaging arrived should be of little surprise biochemical information provided by means of spectroscopy has no real clinical antecedents, and we must look to the biochemistry research literature for its interpretation. Clinical MR spectroscopy {is} thus relatively uncharted territory in the field of radiology (p. 1). Increasingly, however, modern systems have localization capabilities that can connect imaging to spectroscopy. Therefore, it can be expected that MRSI will very soon establish these clinical antecedents, especially in relation to the advances made in our knowledge of the cell biology of cancer.

1.3 Objectives, scope and organization of this book

This book is written first and foremost for clinicians and clinical researchers at various levels of training, including medical students. Obviously, this book is not a substitute for a textbook in oncology. Nor does its scope allow for a comprehensive exposition of the physics and mathematics that underlie the complexities of magnetic resonance, and which can be found in textbooks. The engineering and technological aspects of the rapidly expanding field of magnetic resonance also can easily fill many volumes. Instead, we have sought herein to offer a text which is self-contained, with ample references on a chapter-bychapter basis to facilitate in-depth study of a given topic. The style is pedagogical, and will be amenable for use in courses on MR as well as in oncology. A glossary of terms used in MRS and MRSI and a very extensive list of acronyms are provided in appendices.

We begin in Part A by providing sufficient background in physics and mathematics for the reader to grasp the basic principles of magnetic

resonance, and to appreciate the strategies that allow MR signals to be transformed into clinically meaningful information. This requires an understanding of signal processing, which is initially presented intuitively, using a conceptual approach to conjugate variables, with familiar clinical examples. This is followed by a succinct mathematical exposition aimed at a broad audience. A brief overview of key technological aspects of MR and the nomenclature is also given. In Part B, the current state of the art of in vivo MR imaging and spectroscopy for cancer diagnostics is presented, starting with general principles, and then systematically reviewing achievements to date for brain tumors, prostate cancers, gynecological tumors, head-and-neck, lymphomas, musculoskeletal, renal, hepatic, gastrointestinal and other tumors. We end Part B with a chapter on the current approach to screening and early diagnosis of breast cancer, highlighting what MRI and MRS have offered to date. The chapters in Part B include a brief background of epidemiological and clinical aspects of the tumor types examined with MRS/MRSI. Within this background, emphasis is placed upon etiology and risk factors, in order to suggest when the clinical index of suspicion should be raised, and when, therefore, the information provided by MRS and MRSI might be of particular benefit. Other diagnostic modalities and typical findings using MRI are presented, as well, so that the reader will have a comprehensive appreciation of how these tumors are currently identified. The differential diagnosis is thoroughly discussed, which helps the reader to learn about other indications for MRS and MRSI. We also review available in vitro MRS data for each of the malignancies presented in Part B. Such data consistently yield deeper insights into the metabolite features of specific cancers, and help motivate the quest, outlined in the next, and final, part of the book, to extract further information in the clinical setting from molecular imaging via MRS. In Part C, we begin by discussing the current limitations of in vivo MRS/MRSI and their relationship to reliance upon the conventional Fourier-based framework for data analysis. We then demonstrate how recent advances in signal processing via the fast Pad transform (FPT) can circumvent many of these problems. Salient illustrations are then provided. We examine how these advances could impact upon cancer diagnostics, with a view to early detection and screening, including surveillance of high-risk groups. In the concluding chapter, the role of molecular imaging through MR is explored within a broader context, in which working and general quality of life, treatment, clinical monitoring and surveillance are part of an integrated whole.

Our hope is that this book will be part of the overall efforts via translational research to help clinicians more actively and confidently engage in more basic areas such as molecular imaging through magnetic resonance, as part of the battle against the scourge of cancer.

References
[1] J. Tepper, Form and function: The integration of physics and biology, Int. J. Radiation Oncology Biol. Phys. 47, 547-548 (2000). [2] T. Jones, The spectrum of medical imaging, Eur. J Cancer 38, 2067-2069 (2002). [3] National Cancer Institute, Development of Novel Imaging Technologies (Phased Innovation Award), 2000, (http://www.health.gov/healthypeople/). [4] P.G. Kluetz, C.C. Meltzer, V.L. Villemagne et al., Combined PET/CT imaging in oncology. Impact on patient management, Clin. Positron Imaging 3, 223-230 (2000). [5] L.P. Adler, G. Bakale, Positron emission tomography imaging, in: I. Khalkhali, J.C. Maublant, S.J. Goldsmith, Nuclear Oncology: Diagnosis and Therapy Lippincott Williams & Wilkins, Philadelphia, 2001, p. 289-295. [6] R. Hustinx, F. Benard, A. Alavi, Whole-body FDG-PET imaging in the management of patients with cancer, Semin. Nucl. Med. 32, 35-46 (2002). [7] A.E. Li, D.A. Bluemke, Magnetic Resonance Imaging, In: V.T. DeVita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 669-679. [8] S.J. Vaidya, G.S. Payne, M.O. Leach, C.R. Pinkerton, Potential role of magnetic resonance spectroscopy in assessment of tumor response in childhood cancer, Eur. J. Cancer 39, 728-735 (2003). [9] P.A. Bottomley, The trouble with spectroscopy papers, J. Magn. Reson. Imaging 2, 1-8 (1992).

Part A BRIEF OVERVIEW OF BASIC PRINCIPLES

Chapter 2

Magnetic Resonance
_______________________________________________________________________________

Basically, nuclear magnetic resonance techniques, from which MRI and MRS sprang, involve manipulation of nuclear magnetic dipole moments1 by means of externally applied magnetic fields, and the subsequent recording and mathematical analysis of radio signals emitted from the nuclei in response to these manipulations. Magnetic moments are vectors associated with vector angular moments or spins. Spin is a purely quantum-mechanical property of matter. This should be distinguished from mechanical spin, e.g. of a spinning top. A proton rotates around its own axis and this produces a spin, which is a vectorial quantity whose projections are quantized. In other words, quantum mechanics predicts that only a certain limited number of projections of a spin on a given axis are allowed, and this has been confirmed experimentally. As such, spins cannot be explained by classical physics. Similarly, magnetic moments are also quantized. Resonance is defined as constructive coherence of waves, of the same frequency and phase, leading to an increase in amplitude. The resonance frequency is the frequency at which this process occurs.
1 Only certain nuclei with an odd mass number possess a net magnetic dipole moment for the nucleus as a whole. These include: 1Ha single proton, the most ubiquitous. Others are 13 C, 23 Na, 31P, 19F (as a molecular label, e.g. of drugs or metabolites), and these can produce magnetic resonances. More accurate notation for these nuclei would be 1H+, 13C6+, 23N7+, 31P15+, 19F9+ where the charge states are shown. Nevertheless, following the common practice, we shall use the labels 1 H, 13C, etc.

This physical phenomenon within magnetism was discovered by I.I. Rabi et al. [1] in 1938.

2.1 Precession: Response of 1H and some other nuclei to a static external magnetic field
Without external influences, nuclear magnetic moments are randomly oriented in space. When these designated nuclei are exposed to a static external magnetic field, the Lorentz force will act on their magnetic moments so as to orient them in parallel or anti-parallel to the external field. Because of the nucleis spins, the process of alignment with the external field occurs in circular movements, called Larmor precession, as illustrated in Figure 2.1. On appearance only, this is reminiscent of a spinning top, but cannot be reduced to it [2]. Figure 2.1 Precession response of a spinning nucleus to a static r magnetic field of strength B0 .

Precession

Magnetic moment Spin Spin

r B0
The plane of precession is indicated by the dotted line, as the spinning nucleus progressively aligns itself with the static magnetic field

r B0

The magnetic dipole moment (or magnetization) of the precessing nucleus has a magnitude and a direction, and can therefore be expressed by a vector (Figure 2.2). Figure 2.2: Magnetization vector of the precessing nucleus in response to a static, external magnetic field ( B0 ).

Mz

r B0
Mxy= the transverse component, Mz= the longitudinal component, and full line) is the net resultant magnetization vector

Mxy

r M

(indicated with heavy,

The frequency of precession of M around B0 is called the Larmor frequency L and is dependent upon both the strength of the magnetic field B0 in units of tesla (T) and the gyromagnetic ratio , which is constant for a given nucleus, in units of megahertz per tesla (MHz/T). The Larmor relationship is expressed as: L = B0 . (2.1)

The Larmor frequency L is the angular frequency, and is related to the linear frequency as follows: L = 2 . (2.2)

10

For proton hydrogen, = 42.576 MHz/T. Thus, with an external magnetic field of 1.5T (as is most often applied in clinical scanners), the Larmor frequency of proton hydrogen is calculated to be: L = (42.576 MHz/T)(1.5T) = 63.864 MHz. Parallel and anti-parallel orientations When exposed to an external field B0 , the spin of the nucleus may be at one of two discrete energy (E) levels, according to the principles of quantum mechanics. At the low spin-energy level, the longitudinal component of the magnetic vector is parallel to the magnetic field, and at the high energy level it points in the opposite direction (anti-parallel), as shown in Figure 2.3. In bulk matter, the number of e.g. protons with parallel magnetic moments is only slightly larger (by about a factor of 10-4) than that of protons with anti-parallel magnetic moments. Figure 2.3: Parallel and anti-parallel orientations of precessing nuclei r in response to an external static magnetic field B0
(Adapted from Fleckenstein et al. [2] and from Farr et al. [3]).

(2.3)

Low E State

r B0

High E State

The heavier lines and darker color indicate the greater stability of the low energy state.

11

Net Magnetization Vector The difference between the high and low energy state is the net magnetization of the nucleus within the field. Since the higher state is less stable, there will always be a net magnetization in the parallel direction (Figure 2.4). Note that the transverse vectors cancel because their phases are random and the static magnetic field is incapable of cohering them. Figure 2.4: Net magnetization vector parallel to the static magnetic field (Adapted from Fleckenstein et al. [2] and from Brown et al. [4]).

r B0
Mz

Note slight excess of protons (10-4) in the low spin-energy level. Transverse vectors cancel because the phases are random and, as such, cancel out. The heavy black line indicates the net magnetization vector

2.2 Response to 90o radio-frequency pulse: Resonance and relaxation

After exposure to a static magnetic field B0 , the above-described equilibrium state with a net magnetization vector M is achieved within seconds. This can be disturbed and shifted by an externally applied pulse of electromagnetic waves at the Larmor frequency. Only radiofrequency (RF) waves of precisely this frequency (energy) will transfer

12

energy by resonance to the precessing protons2. This is done with an r RF magnetic field oriented perpendicularly to the main field B0 (Figure 2.5).

Figure 2.5: 90 radio frequency pulse (2nd magnetic field) at the r Larmor frequency, in addition to the static magnetic field B0

r B0

RF pulse

This 90 RF pulse forces the transverse components M xy of the proton magnetization to precess in phase in the XOY (transverse) plane, which r r is perpendicular to B0 . This single resultant M xy is large and emits a strong radiosignal at the Larmor frequency (Figure 2.6).

Throughout this book we shall use the terms energy and frequency as synonyms with the convention E = h = , since h = 1, recall that h = h (2 ), where Planck' s constant is denoted by h.

13

Figure 2.6: 90 radio frequency pulse has now flipped the longitudinal vector onto the transverse plane and has forced the transverse components to precess in phase. The resultant Mxy vector sends out a strong radiosignal at the Larmor frequency

r B0

Mxy

2.2.1

Free induction decay

When the RF pulse is turned off, the excited nuclei return to their initial equilibrium state. This process of relaxation generates a signal, which is picked up by a receiver coil placed around the investigated part of the body. Stated equivalently, the time signal is created by the very process of returning of the net magnetization vector from the transverse plane to the equilibrium orientation. The intensity/magnitude of this vector is an exponential function of time. The state of the nucleus associated with this vector is unstable and, as such, decays with time. This is why the time signal is called the free induction decay (FID),

14

reflecting the loss of phase coherence after interruption of the RF pulse. An example of an FID as seen in a typical time signal from a clinical MR scanner is shown in Figure 17.1. The term FID is based upon the following definitions: Free Occurring after the dynamic external perturbations have ceased, 3 Induction Small current from relaxation of proton magnetization r ( ) in the coils surrounding the scanned organ, Decay Decreased probability of occurrence with time: in this case the current, which is converted into a time signal c(t) as an envelope which decays exponentially over time. The FID curve, or equivalently, the time signal c(t) is a heavily packed, oscillating function whose informational content cannot be visually discerned in the time domain, and is therefore transformed into the frequency domain by certain, well-established mathematical procedures as discussed in Chapter 4. 2.2.2 T1 and T2 relaxation, proton density and weighting

Relaxation is comprised of two distinct processes, both of which occur exponentially, but are independent of each other and have different time constants (T1 and T2). The T1 relaxation reflects nuclei falling back from the high to the low energy state, where T1 (the spin-lattice relaxation time constant) is the time for 63% recovery of the longitudinal magnetization (Mz). The T2 relaxation reflects the decay of transverse magnetization (Mxy) due to loss of phase coherence among the precessing nuclei (spin-spin relaxation), where T2 denotes the time at which the transverse magnetization has decayed to 37%4 of its maximum strength. T1 relaxation is always longer than T2 relaxation. The strength of the tissue signal also depends upon the proton density (PD), i.e. the number of hydrogen nuclei present per unit area of tissue.

3 4

But not the static magnetic field, which is constantly present during scanning. This comes from the differential decay law, where the mean life of the metabolite is defined as e-1 x 100% = 37% of the original value.

15

T1 and T2 relaxation and proton density of fluids, water and fat-based tissues Fluids have the highest PD (> 95%), whereas water- and fat-based tissues have similar PD (68 85%)[5]. Fluids have long T1 relaxation times (about 1500 2000 ms), tissues that are water-based have intermediate T1 relaxation times (400 1200 ms), while fat-based tissues usually have short T1 relaxation times (100 150 ms). Water bound to the surface of proteins also has a short T1 relaxation time [3, 5]. Fluids also have the longest T2 relaxation times (700 1200 ms). The T2 relaxation times are usually between 40 and 200 ms for water-based tissues, and 10 100 ms for fat-based tissues [5]. Generally, the greater the percentage of free water in a given tissue, the longer the T2 relaxation time [3]. Parameters that are adjusted to achieve weighting The fact that T1 and T2 relaxation times differ among various tissues is the basis for creating contrast in MRI [5, 6]. Weighting is accomplished by adjusting the following parameters in an MR recording (encoding) sequence.
Repetition time (TR) The repetition time is defined as the time between successive 90RF pulses. A longer TR allows more of the radio frequency energy to be dissipated by the proton through spin-lattice relaxation, and thereby produces less T1 weighting [4, 5]. Echo time (TE) The signal, which comes back from the patient, is collected as an echo. The echo time is the time from the 90 RF pulse to the echo (signal) maximum. Longer TE provides more time for proton dephasing and thereby produces more signal from tissues with long T2 values [4, 5].

Types of weighting
T1 weighted images T1 weighting is achieved with a short TR and short TE. Fat appears very bright on T1 weighted images, because fat has a rapid T1 relaxation, while water-based tissues are mid-grey and fluids are very dark, unless they are fast moving. As the TR is lengthened, fat loses its brightness. T1 weighted images show sharp boundaries between different tissues, and therefore have been termed anatomy scans [2, 5, 6].

16

T2 weighted images T2 weighting is achieved with a long TR and long TE. Fluids will have high intensity, while water-based tissues and fat will be of intermediate intensity. As the TE is shortened (keeping TR long to allow full T1 relaxation), fluids progressively lose their brightness. Since collections of abnormal fluid appear bright compared to many of the darker normal tissues, T2 weighted images have been called pathology scans [2, 5, 6].

Table 2.1 provides some examples of signal intensities seen with various tissues using T1 and T2 weighting. Tissues that appear bright on MRI are said to have high signal intensity [6]. Table 2.1 Relative signal intensities of various tissues with T1 & T2 weighting
(From Refs. [3 & 5-7])

T1 weighting Brain Cerebrospinal fluid Grey matter White matter Liver Muscle Pancreas Spleen Prostate Central Zone Peripheral Zone Uterine Endometrium

T2 weighting

Dark Intermediate to dark Bright Intermediate to bright Dark Intermediate to bright Intermediate

Very bright Intermediate to bright Intermediate to dark Intermediate to dark Dark Intermediate to dark Bright

Intermediate Intermediate Intermediate

Dark Bright Bright

PD weighted images PD weighted images have less contrast than T1 or T2 weighted images. This is because the proton densities, which reflect water content, are fairly similar for various tissues. Both fat and water appear with intermediate signal intensity with PD weighting. Grey matter has a somewhat greater proton density than

17

does white matter [3]. PD weighted imaging is applied in certain clinical situations, such as for visualization of nerve roots in the cervical spine [5].

2.2.3

Spin echo versus gradient echo sequences

Spin-echo pulse sequence The spin-echo pulse sequence is considered the gold-standard for MRI. Here, an initial 90 RF pulse is followed by a 180 RF pulse that acts to rephase the nuclei and thereby to generate an echo, which is the signal that is detected by the receiver coil. Gradient-echo pulse sequence The initial RF pulse produces a flip-angle which is < 90, and is rephased by a magnetic gradient to generate an echo signal (see Chapters 4 & 5). Shorter TR and TE can be used and this allows the scan to be obtained much more rapidly. T1, T2 or PD weighting can be applied with gradient-echo pulses, but are more influenced by inhomogeneities of the main magnetic field. Gradient-echo pulse sequences have been particularly useful for MR angiography. T1 weighted gradient echo images can be acquired very rapidly, and can be useful for 3D volume scanning, for breath-hold chest or abdominal imaging [4,5].

2.3

Technical Considerations

MR scanning is performed within a Faraday cage to minimize interference from the surroundings. 2.3.1 The components of a clinical scanner The main magnet This is a superconducting magnet, which produces a very strong and homogeneous field inside the bore (the opening where the patient lies). The magnet must be extremely stable in time, i.e. it must be static. Most clinical scanners currently operate at 1.5T. Note that a stronger magnetic field yields improved signal-to-noise ratio (SNR), but homogeneity is crucial to obtain the benefit of this increased field strength. Superconducting magnets require liquid helium as a cryogenic cooling fluid [5].

18

Gradient Coils These are situated inside the bore of the main magnet, and are used to produce magnetic field gradients in the X-, Y- and Z-axes. These can be rapidly varied over time. The gradient magnetic fields are needed to allow the signal to be localized to precise positions within the patient (see Chapters 4 & 5). The use of gradients also allows MR images to be acquired in multiple planes, including coronal, sagittal, axial and oblique [6]. Radio frequency transmitter/receiver coil The transmitter RF coil surrounds the body or site of interest. A body coil is usually built into the magnet. The receiver coil detects the MR signals produced by the body. Shielding from the Faraday cage is needed to minimize interference between these relatively weak signals and other external fields. Besides the body and head coil, a number of specialized coils have been developed, e.g. breast, endocavitary, knee, extremity coils, etc. These specialized coils are smaller and fit the anatomy more closely, thereby improving image quality. Coil layout can also greatly impact upon final image quality [4,6]. The gantry The couch or gantry upon which the patient lies is located in the centermost part of the scanner. Besides providing the possibilities to monitor and adjust the patients position, equipment for physiological monitoring (peripheral pulse, ECG, respiration) may be included [5]. 2.3.2 Motion Artefacts

One of the best overall strategies to reduce motion artefacts is to reduce the scan time as far as possible. This is one of the reasons for the use of gradient-echo pulse sequences, which enable very rapid image acquisition [8]. Gross patient motion Careful preparation of the patient is of utmost importance, with maximum avoidance of patient discomfort. Pads or immobilization equipment are sometimes employed. Sedation may be necessary in carefully selected patients for whom the scan is indispensable and who by themselves cannot otherwise restrain their gross motor activity.

19

Physiological motion
Respiratory motion If the scan time can be reduced to 25 seconds, the scan can be obtained during a single breath hold, or via a series of breath holds. Insofar as breath-holding is not feasible, the number of signal acquisitions can be increased, or respiratory gating can be employed, but the latter prolongs the TR to such an extent that T1 weighted images cannot be obtained, and T2 or PD weighted images require inordinately long encoding times. Moreover, gating depends upon a regular breathing cycle. Other alternatives include: Respiratory-Ordered Phase Encoding (ROPE), which does not prolong scan time, but cannot be used with fast spin-echo sequences. Navigator echoes that can compensate for various types of motions with high resolution, but these are not yet widely available on clinical scanners. For more details see Ref. [5] and references cited therein. Cardiac motion Cardiac motion artefacts are removed by gating the sequence to the cardiac cycle, with the peak of the R wave detected by the scanner and used to trigger the imaging sequence. Cardiac gating can also be accomplished via digital arterial plethysmogram. Peristaltic motion Anti-peristaltic medications such as hyoscine butylbromide or glucagon have been used, or ultrafast sequencing.

References
[1] I. Rabi, Z.R. Zacharias, S. Millman, P. Kusch, A new method of measuring nuclear magnetic moment, Phys. Rev. 72, 318 (1938). [2] P. Fleckenstein, J. Tranum-Jensen, Anatomy in Diagnostic Imaging, Blackwell, Copenhagen, 2001. [3] R.F. Farr, P.J. Allisy-Roberts, Physics for Medical Imaging, W.B. Saunders, Edinburgh, 2000. [4] M.A. Brown, R.C. Semelka, MRI basic principles and applications. 2nd Edition, John Wiley & Sons, New York, 1999. [5] D.W. McRobbie, E.A. Moore, M.J. Graves, M.R. Prince, MRI from picture to proton, Cambridge University Press, Cambridge, 2003. [6] A.E. Li, D.A. Bluemke, Magnetic Resonance Imaging in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 669-679.

20

[7] C. Reinhold, I. Khalili, Postmenopausal bleeding: value of imaging, Radiol. Clin. N. Am. 40, 527-562 (2002). [8] P. Mansfield, I.L. Pykett, P.G. Morris, Human whole body line-scan imaging by NMR, Br. J. Radiol. 51, 921-922 (1978).

21

Chapter 3

In vivo Magnetic Resonance Spectroscopy

_______________________________________________________________________________

3.1 Chemical shift as the basis for identifying metabolites via a frequency spectrum
Thus far, we have been discussing the processes common to both MRI and MRS. The key to MRS is that the resonant frequency varies slightly for protons or other nuclei displaying net magnetic dipole moments, because the electron cloud surrounding the nucleus differs for various compounds. In other words, nuclei resonate at slightly different frequencies, depending upon the reduction of their intrinsic magnetic fields by the shielding effects from the surrounding electronic clouds. These small differences in resonant frequencies are the basis of the phenomenon called chemical shift, from which a chemical shift frequency spectrum of metabolites is obtained. Thus,

p = L (1 + )

(3.1)

where is the screening constant representing chemical shift (dimensionless), p is the actual frequency of e.g. proton magnetization r r returning to alignment with B0 , and L is the Larmor frequency. Although is a small correction ( <<1), it nevertheless represents the basis of MRS.

22

Dimensionless units: parts per million irrespective of magnetic field strength The chemical shift or frequency spectrum is labelled in dimensionless units of parts per million (ppm)1 rather than in hertz (Hz) (cycles/sec). The reason for this convention is so that all spectra are standardized, irrespective of the magnetic field strength B0. Otherwise, since the Larmor frequency L from Equation (2.1) is proportional to B0, a given metabolite registered at two different field strengths would appear at unequal positions in a spectrum. Thus, e.g. while water and fat resonate approximately 210 Hz apart with a 1.5T magnet, the peaks are 420 Hz apart at 3.0T. Converting to ppm, we see that 210 Hz at 1.5T yields 210 Hz/63.9 MHz = 3.3 ppm, and that 420 Hz at 3.0T yields 420 Hz / 127.6 MHz =3.3 ppm. Conversion from Hz to ppm is achieved as follows:

[Observed frequency (Hz)] - [Reference frequency (Hz)] . (3.2) Larmor Frequency (MHz)
For proton MRS, the compound tetramethylsilane (TMS)2 is used as a reference value of 0, such that:

ppm =

Observed frequency (Hz) Observed frequency (Hz) = . (3.3) Larmor Frequency (MHz) [Larmor Frequency (Hz)] 106

A 1H chemical shift spectrum computed in Ref. [1] using the FID encoded with a 4T magnet from the brain occipital grey matter of a healthy volunteer is shown in Figure 3.1. The time signal was measured by the group at the Center for Magnetic Resonance, University of Minnesota [2].

1 2

ppm refers to the fractional shift in frequency relative of the center frequency of the magnet. TMS is used in NMR spectroscopy in chemistry and is not found in the body

23

Figure 3.1 In vivo 1H MR magnitude spectrum at 4T from the occipital grey matter of a health human volunteer. From Ref. [1], with permission.

Note that the abscissa (chemical shift) is read from right to left.

MR visible compounds Currently, only a fairly small number of compounds are readily observable on a clinical scanner (usually 1.5T), using conventional data processing methods. These compounds must be present in high concentrations ( 1 2 mM) and must be small and mobile (with T2 relaxation times 10 20 ms). It should be noted that large macromolecules have T2 relaxation times in the sec range [3-6].

24

3.2

Characteristics of the metabolites evaluated by MRS

Here, the basic characteristics of the metabolites evaluated by MRS are presented to provide the reader with an initial intuitive grasp of their meaning and importance. These will be discussed in more depth in Chapter 4 and thereafter. In Chapter 4 we will explain how an MRS time signal is converted into a frequency spectrum.

Frequency

Frequency denotes the position of the peak, determined by the type of nucleus and the chemical environment (chemical shift). As explained in Section 3.1, rather than being expressed in units of hertz, frequency is expressed in dimensionless units of ppm, irrespective of the static magnetic field strength.

Amplitude

The amplitude is the intensity of the signal at a given frequency, and is determined by the concentration of the metabolite. This is usually measured in arbitrary units. The peak area can provide a more precise estimate of metabolite concentration. This depends not only upon amplitude, but also fullwidth at half-maximum amplitude (FWHM), which is inversely related to T2 relaxation time (see later chapters, and especially Part C, for a more detailed discussion of quantification, and its attendant difficulties).

Relaxation /decay characteristics

Each metabolite has a characteristic T2 relaxation time, and these differ markedly. At longer echo times, many metabolites will have already decayed. As a consequence, fewer peaks will appear, their amplitudes will be diminished, and ratios of various metabolites will also be altered [3].

Phase

Besides amplitude and frequency, a sinusoidal signal or waveform also has a phase, which describes the instantaneous position within the cycle. This can range from 0 to 360, and can be used to help determine the spatial position [7].

25

J coupling

J coupling is an interaction between nuclei of adjacent atoms, which causes a splitting of the resonance peak into two or more peaks, i.e. doublets or multiplets. This interaction is between or among spins (spin-spin interaction). J coupling also causes phase evolutions that can lead to peak and baseline distortions. These distortions can vary with echo time and field strength [4,7].
1

3.3

H (proton) MRS

As noted in Chapter 2, several nuclei with an odd mass number possess a net magnetic dipole moment for the nucleus as a whole, and these can produce magnetic resonance. Of these, the hydrogen atom nucleus consisting of a single proton is the most ubiquitous. For this reason, plus having the most favorable magnetic sensitivity (gyromagnetic ratio) yielding the highest SNR, proton hydrogen or 1H has been the most widely used for MRS. An additional advantage of 1H MRS is that it can be performed with the same equipment (RF coils, amplifiers, etc.) as is used in standard MRI [5]. 3.3.1 The in vivo proton MR spectrum

The in vivo proton MR spectrum varies according to the tissue being examined, relevant aspects of which will be discussed in Part B. Here we will briefly review some of the major MR visible metabolites found in the normal brain. The reader is referred to Ref. [3-5, 8] for more details about the biochemical and MR characteristics of these metabolites. As seen in Figure 3.1, these include:
N-Acetylaspartate (NAA) [2.0 ppm] whose acetyl moiety (CH3) resonates at 2.0080 ppm. NAA is present only in living neurons, and is a marker of neuronal density and viability. It is the largest peak in the normal brain, after water suppression. Creatine (Cr)[3.0, 3.9 ppm] whose N(CH3) moiety resonates at 3.03 ppm, and whose CH2 moiety resonates at 3.9 ppm. Creatine reflects total creatine stores in cells and plays a primary role in maintaining the energy storage systems in cells. It is the sum of creatine + phosphocreatine (PCr).

26

Choline (Cho) [3.2 ppm] whose N(CH3)3 moiety resonates at 3.2 ppm. Choline reflects phospholipid metabolism of cell membranes. It is a marker for membrane damage, cellular proliferation and density. Total choline is the sum of phosphorylcholine + glycero-phosphorylcholine + 5% free choline. Choline is used as a key spectroscopic indicator of malignancy. Only a small part of the MR visible choline is from the neurotransmitter acetylcholine. Lactate (Lac) [1.3 ppm] whose CH3 moiety resonates at 1.31 ppm. Lactate is an indicator of anaerobic metabolism, with low concentrations seen in the normal brain. At intermediate TE ( 140 ms), lactate peaks invert below baseline and appear as a doublet due to J coupling. Myoinositol (mI)[3.52, 4.05 ppm] whose (CH)4 moiety resonates at 3.52 ppm and CH moiety at 4.05 ppm. Myoinositol has a number of roles: cell-volume regulator (osmolyte) in astrocytes, as well as being a myelin degradation product. Myoinositol has a short T2 relaxation time, and therefore can be detected only at short echo times. Glutamine, glutamate (Glx)[2.02, 2.5, 2.63, 3.8 ppm] whose peaks are low, despite high concentrations due to J coupling. Glx requires a short TE (30-35 ms) to be identified. Glx is an astrocyte marker, glutamate is an excitatory neurotransmitter and glutamine is involved in regulating neurotransmitter activity and in detoxification processes.

3.3.2

Suppression of giant resonances

Water Suppression The proton resonance from water is about four orders of magnitude larger than any other metabolite resonances. In order to obtain information about these metabolites, water suppression must first be performed. This can be achieved by applying a very narrow frequency-selective pulse (called CHEmical Shift Selective (CHESS) pulse) precisely at the Larmor frequency of water, 4.7 ppm. This is followed by gradient pulses to spoil any transverse magnetization. The CHESS pulses must be very accurate in order to optimally suppress water and avoid suppressing neighboring peaks [3, 4, 7]. Unfortunately, many published data encoded at clinical scanners with 1.5T exhibit too large residuals after water suppression. This causes severe deformations of spectral features of metabolites near 4.7 ppm. Lipid Suppression In proton MRS performed outside the brain, lipid signal is also of much larger magnitude than the metabolites of interest [9]. This is

27

particularly problematic, e.g. for the breast, which has an abundance of mobile lipid. Techniques used for lipid suppression include averaging spectra over different echo times [10], application of long echo times (which leads to loss of signal intensity) [11], inversion recovery technique [12], etc.

3.4

31

P MRS

In comparison to 1H MRS, 31P MRS has the following advantages and disadvantages: Can detect a number of clinically important compounds P MRS has been used to detect a number of clinically important compounds. These include, most notably, adenosine triphosphate (-, - and -ATP, which resonate respectively at approximately [16.5 ppm, -7.5 ppm, -2.5 ppm]), phosphocreatine (used as the reference chemical shift at [0 ppm], inorganic phosphate (Pi) [5 ppm], the phosphodiesters around [3 ppm] and phosphomonoesters around [7 8 ppm]. The pH can also be calculated from the chemical shift. These have been informative for clinical oncology, as will be reviewed in Part B. Other applications of 31P MRS include assessment of muscle metabolism during exercise, muscular dystrophy, cardiac energetics, and hepatic abnormalities [7]. Does not require water or lipid suppression A major advantage of 31P MRS is that it does not require water or lipid suppression. Lower gyromagnetic ratio lower signal Phosphorus has a lower Larmor frequency than hydrogen. At 1.5T phosphorus resonates at 25.9 MHz. Its gyromagnetic ratio is 17.235 MHz/T compared to 42.576 MHz/T for protons. Since the signal strength is proportional the gyromagnetic ratio, 31P MRS generates a much lower signal compared to 1H MRS. Phosphorus is less abundant lower signal Phosphorus is also much less abundant than protons. Its whole body concentration is about 1000 lower than that of proton hydrogen. For the above two reasons, the SNR with 31P MRS is much poorer than for 1H MRS. For example, it was estimated that the signal from a lesion
31

28

such as a tumor with 31P MRS would have to be nearly 10 times stronger to obtain the same SNR as with 1H MRS [11]. Requires a second RF subsystem Unlike 1H MRS, which can be performed with the same equipment (RF coils, amplifiers, etc.) as is used in standard MRI, 31P MRS requires a second RF subsystem [5, 7]. Requires special sequences A special type of sequence that involves encoding the FID immediately after excitation is required for 31P MRS (see Section 3.5). This is much more time consuming compared to 1H MRS. Requires proton decoupling Weak interactions between phosphorus atoms and protons in the molecules or in surrounding water lead to a broadening of spectral lines. By proton decoupling, 31P spectrum will contain sharper lines and proportionally high peak intensities (given that the peak area is constant).

3.5

MRS of a single volume

With single volume MRS, the MR signal is obtained from only one site within the tissue or organ. This site is called the volume of interest (VOI) and is the intersection of three orthogonal planes. The volume or voxel3 is usually about 1.5 to 4 cc. Slice selective radiofrequency pulses are used to measure the spectrum within the voxel [3]. 3.5.1 Voxel positioning

Proper positioning of the voxel is vital to obtaining an MR signal which is of good quality and which provides diagnostic information. As will be discussed in Chapter 5 and in Part B, it is often difficult to choose a single voxel which reflects the underlying pathology, e.g. tumor, as this can be mixed with necrosis, edema, and even normal tissue. It is important to try to choose a voxel which is free from blood and blood products, necrosis, metal, calcifications and bone. Otherwise, differing magnetic susceptibility occurs, and this results in a non-homogeneous field that diminishes the diagnostic quality of the MR spectra [13].

The terms volume of interest and voxel of interest are often used interchangeably [3].

29

3.5.2
1

Single voxel sampling methods

H MRS

STEAM (STimulated Echo Acquisition Mode) Uses three selective 90 RF pulses, each with a gradient alongside one of the three axes, to acquire the stimulated echo from the voxel. STEAM has lower SNR than PRESS or ISIS (see next single-voxel techniques), but requires shorter echo times. Can detect more complicated spin systems such as glutamate and glutamine, and short-lived metabolites like myoinositol. STEAM does not show the inversion of lactate at TE 144 ms. With STEAM there is incomplete recovery of the signal [3, 4, 7, 13]. PRESS (Point-RESolved Spectroscopy) Uses one 90 RF pulse and two 180 RF pulses to obtain a spin echo. Each of these three pulses has a slice-selective gradient along one of the three main axes. PRESS provides improved SNR compared to STEAM and is less sensitive to diffusion and motion. In the past, PRESS required longer echo times. At present, with improved radiofrequency technologies, the 180 RF pulses can be generated more cleanly, and PRESS can be performed at short TE. With PRESS, there is complete recovery of the signal [3, 7, 13].

Because the SNR is twice greater with PRESS compared to STEAM and now that short TE can be used, PRESS is the preferred single-voxel sampling method for MRS [7, 13].
31

P MRS

Since in vivo 31P MRS has a much shorter T2 compared to 1H MRS, echo-based sequences such as STEAM and PRESS cannot be applied, even at short TE. Rather, the FID is acquired immediately after excitation.
ISIS (Image Selective In vivo Spectroscopy) Most often used in 31P. Unlike STEAM and PRESS, which can be performed in a single experiment, ISIS requires 8 experiments, with an add/subtract scheme to cancel out signals from regions outside the voxel of interest. As mentioned in Section 3.4, this is time consuming, since eight TRs are required for each signal acquisition [7, 12]. DRESS (Depth RESolved Spectroscopy) This technique uses the properties of the surface coil to excite a tissue at a certain depth. The volumes are large compared to ISIS, and, as a consequence, small lesions, e.g. tumors, will not be detected. Moreover, the MR signals can be contaminated from surface tissue signals [7].

30

3.6

Magnetic field inhomogeneity

3.6.1 Relationship of the spin-echo phenomenon to magnetic field inhomogeneity Loss of phase coherence (dephasing) means loss of RF signal. Part of this loss is due to spin-spin relaxation, expressed by T2, which is an inherent property of the material. The observed rate of decay of phase coherence denoted by T2*, is always faster than T2, because of inhomogeneities of the magnetic field. The part of dephasing due to inhomogeneities can be cancelled by the spin-echo maneuver, in which a 180 RF pulse is applied after termination of the 90 RF pulse. The latter serves to re-gather the transverse relaxation vectors so that they emit a strong signal [14, 15]. 3.6.2 Shimming

Inhomogeneities in the static magnetic field will lead to rapid decay of the FID; the signal amplitude will appear lower than its true value, with widening and distortion of the peaks. A key component of preparation for MRS is to make correction of magnetic field inhomogeneities, by adjusting static gradients. The process of adjusting the field gradients to maximize magnetic homogeneity over the voxel is termed shimming [4, 7]. Shimming can be performed manually or automatically, the latter usually being much faster and based upon phase-sensitive gradient echo images, iterative techniques, etc. [16].
Full-width at half maximum or line-width The FWHM of the water resonance is used to gauge the voxel homogeneity. This is also termed line-width and is expressed either in Hz or in ppm. A line width of 0.08 ppm (5 Hz at 1.5T) is considered ideal [7].

References
[1] Dz. Belkic, Exact analytical expressions for any Lorentzian spectrum in the fast Pad transform, J. Comp. Meth. Sci. Eng. 3, 109-186 (2003). [2] I. Tk, P. Andersen, G. Adriany, H. Merkel, K. Ugrbil, R. Gruetter, In vivo 1H NMR spectroscopy of the human brain at 7 T, Magn. Res. Med. 46, 451-456 (2001). [3] E. R. Danielsen, B. Ross, Magnetic Resonance Spectroscopy Diagnosis of Neurological Diseases, Marcel Dekker, Inc., New York, 1999.

31

[4] D. J. Drost, W.R. Riddle, G.D. Clarke, Proton magnetic resonance spectroscopy in the brain: Report of AAPM MR Task Group #9, Med. Phys. 29, 2177-2197 (2002). [5] D. Spielman, In vivo proton MR spectroscopy: basic principles and clinical applications, Section for Magnetic Resonance Technologists Educational Seminars 3, 19-37 (2000). [6] K. Belkic, Magnetic resonance spectroscopy and spectroscopic imaging: a review of basic principles and achievements in oncology, J. Comp. Meth. Sci. Engin. 3, 505-533 (2003). [7] D.W. McRobbie, E.A. Moore, M.J. Graves, M.R. Prince, MRI from picture to proton, Cambridge University Press, Cambridge, 2003. [8] V. Govindaraju, K. Young, A.A. Maudsley, Proton NMR chemical shifts and coupling constants for brain metabolites, NMR Biomed. 13, 129-153 (2000). [9] R. Kreis, 1H MR spectroscopy outside the brain, MAGMA 15 (Suppl. 1), 2 (2002). [10] P.J. Bolan, L. DelaBarre, E.H. Baker, et al. Eliminating spurious lipid sidebands in 1H MRS of breast lesions, Magn. Reson. Med. 48, 215-222 (2002). [11] R. Katz-Brull, P.T. Lavin, R.E. Lenkinski, Clinical utility of proton magnetic resonance spectroscopy in characterizing breast lesions, J. Natl. Cancer Inst. 94, 1197-1203 (2002). [12] B. Vikhoff-Baaz, In vivo MRS and MRSI: Performance analysis, measurement considerations and evaluation of metabolite concentration images, Gteborg University, [Doctoral Dissertation], 2000. [13] L.A. Brando, R.C. Domingues, MR Spectroscopy of the Brain, Lippincott Williams & Wilkins, Philadelphia, 2004. [14] P. Fleckenstein, J. Tranum-Jensen, Anatomy in Diagnostic Imaging, Blackwell, Copenhagen, 2001. [15] R.F. Farr, P.J. Allisy-Roberts, Physics for Medical Imaging, W.B. Saunders, Edinburgh, 2000. [16] R. Kreis, Quantitative localized 1H MR spectroscopy for clinical use, J. Prog. NMR spectroscopy 31, 155-195 (1997).

33

Chapter 4

Signal Processing for Magnetic Resonance: The Basics

_______________________________________________________________________________

4.1

The concept of conjugate variables: Complementary representations

A critical concept to understanding both MRS and MRI is that of conjugate variables, which allow for complementary representations. Many times, it occurs that signals are recorded in such a way that it is very difficult or even impossible to interpret or disentangle them as they stand. One such example is a free induction decay (as seen, e.g. in Figure 17.1). When looking at an FID, we see a heavily packed, oscillating function, from which not even the trained eye of the most astute clinician could decipher much meaningful information. Yet, it is from the FID that the MR spectrum is obtained. The key is that the information contained in the FID, which is a signal recorded over time (abscissa), is transformed into its complementary representation, i.e. the frequency domain, and the abscissa then becomes frequency. This is possible because time and frequency are conjugate variables, such that their representations can be mapped with fidelity from one to the other via the exact Fourier integral, the mathematical details of which are briefly described in Section 4.4. Another pair of conjugate variables is momentum and position, such that the momentum representation and the coordinate representation can also be mapped from one to the other via the exact Fourier integral. This mapping is the basis for engendering an MRI image in coordinate

34

space from the recorded signal, which is obtained in the momentum or k space. Simply stated, it is often desirable to map data from a given domain to its complement in order to access the information contained in the measured data, but which is virtually uninterpretable in the form in which it was originally measured. Thus, for example, the FID is experimentally measured, while the meaningful, quantifiable information is provided by the theory in signal processing analysis.

4.2
4.2.1

From the time to the frequency domain

Transformation from the time to frequency domain in medicine: Some familiar clinical examples outside MRS

In medicine a number of familiar examples can be cited in which time signals are transformed into the frequency domain in order to gain important clinical insights. In neurology, we know that slow activity (theta 4-7 Hz or delta < 4 Hz) may not always be apparent upon visual inspection of the recorded electroencephalogram (EEG). Theta and delta activity can be not only identified, but also quantified via spectral analysis. In cardiology, late potentials at 60 120 Hz may occur just after the QRS complex1. These cannot usually be visually deciphered on the electrocardiogram (ECG). To be identified and quantified, the ECG time signal is transformed into the frequency domain. Another example from cardiology is heart rate variability (HRV). It is well known that severely diminished or absent HRV is an indicator of risk for sudden arrhythmic cardiac death after acute myocardial infarction. Although some appreciation of respiratory sinus arrhythmia can be gained from the ECG time signal, quantification of the 0.15 to 0.4 Hz spectral component requires transforming the recorded ECG data into the frequency domain.

1 The Q R and S waves which together form the QRS complex, representing depolarization of the ventricles of the heart

35

4.2.2

Spectral analysis of the FID: A signal that decays over time

For the above-mentioned clinical examples, visual analysis of the recorded time signal data, which are periodic or quasi-periodic without decay, such as the ECG or EEG, is informative. Transforming these physiological time signals into the frequency domain can be very helpful for certain clinical purposes but, obviously, this is not always required since substantial interpretation is made directly from the recorded signal. In contrast, however, we have noted that the recorded MRS time signal, or FID as is, does not provide clinically meaningful information. Here, transformation into the frequency domain is indispensable if we are to decipher the information contained therein. That information is the relative strengths of the various frequency components, which represent the concentrations of certain MR visible metabolites. The process of separating a signal into its frequency components is known as frequency or spectral analysis. In current practice, this is most often performed using the Fourier transform, in which the signal is multiplied with the sine and cosine functions and the result is summed up for each frequency. As stated, looking at the FID itself does not provide clinically meaningful insights. Thus, in this direct way, we learn nothing about the metabolic constituents of that tissue that has been subjected to manipulations of magnetic fields. But what we do see is that overall the envelop of the FID is an exponentially decaying function over time, and this decay rate can be expressed mathematically as: e-t/T2

(4.1)

If, to take the simplest example, there were only one frequency component present, then the measured FID would be expressed as:

c(t ) = d1e-i1t

(4.2)

where 1 is the complex angular frequency and d1 is the complex amplitude (1 = 1R + i1I , d1 = d1R + id1I with the subscripts R and I denoting the real and imaginary part, respectively). The imaginary part

36

of 1 must be negative in order that c(t) decays to zero as time t increases to infinity (t ): 1 1I < 0.

(4.3)

The exact spectrum F ( ) for a given time function c(t ) is given by the Fourier integral:

F ( ) = dt c(t )ei t .
0

(4.4)

When there is only one frequency 1 with 1I <0, then the exact spectrum is expressed as:

F ( ) = i

d1 1

(4.5)

which is a complex-valued function. The magnitude spectrum

If we take the absolute value F ( ) , we obtain what is called the magnitude spectrum, which is a real quantity. The power spectrum

The power spectrum is the magnitude spectrum squared, and whose maximum is where = 1

F ( ) =
2

d1

1I 4

(4.6)

as shown schematically in Figure 4.1.

37

Figure 4.1 Schematic representation of a power frequency spectrum of a single metabolite peak (a Lorentzian) whose maximum amplitude is at 1R.

F ( )

1R.
The abscissa is frequency and the ordinate is

F ( ) .
2

The quadrature mode: real and imaginary components The FID is detected in the quadrature mode, which produces a spectrum with two sets of data points that are 90 apart in phase. These are the cosine and sine components, which correspond, respectively, to the real and imaginary parts.
Absorption Spectrum Under ideal conditions, i.e. no noise and no initial time delay, the absorption spectrum is the positive-definite real part of the complex spectrum. Otherwise, the absorption spectrum can be constructed from a complex-valued spectrum as e.g. in Ref. [1]. In clinical MRS, the real components of the spectra are mainly used, i.e. the absorption and magnitude spectra. Dispersion Spectrum The dispersion spectrum is the imaginary part of the complex spectrum. This may contain negative frequencies.

38

Phase correction Whenever the initial phase of an FID is not zero, the real and imaginary parts of the complex spectrum contain mixtures of absorption and dispersion mode spectra. Phasing is the process by which the spectrum is sorted into the real and imaginary spectra. Consequences of 1st order phase correction Because of localizing gradients, there is always a delay between excitation and receiving channels, such that the nuclei will dephase by an angle proportional to their frequency. This needs to be corrected by a linear phase shift across the spectrum (first-order phase correction). However, this introduces a slow variation in the baseline, which, in turn, is often corrected with a spline or other polynomial function [2]. A further discussion of phase correction in clinical 1H MRS can be found in Ref. [3].

Consequences of using a finite acquisition time ( T < ) with restriction to the Fourier grid As stated earlier, the exact spectrum is represented by the Fourier integral. The upper limit of the integral is infinity, so that an infinite acquisition time is assumed. In reality, the FID is digitized for only a few tens of milliseconds. If digitization were further extended, mainly noise would be recorded since the FID would have already died away. When using a finite acquisition time, only an approximation of the spectrum can be obtained with the fast (or the discrete) Fourier transformation (see Section 4.4.1). Here, the spectrum is defined only at certain preassigned frequencies or grid points, which are inversely proportional to the total acquisition time (T). By using a finite acquisition time ( T < ) with Fourier-based techniques for signal processing, the shape of the spectrum is distorted, and truncation artefacts appear.
Zero filling A common practice is to add extra data points set to zero, in order to lengthen the signal. Namely, instead of recording a signal with nothing in it but noise, the signal is artificially lengthened by adding zeros. It has been pointed out that this does not improve resolution, but merely renders the shape spectrum more visually presentable [4].

39

Later in this chapter (Section 4.4), we will present some of the mathematical details of the Fourier-based techniques. We will also describe recent advances in signal processing (notably the fast Pad transform) that can overcome this and many other limitations of Fourier-based techniques.

4.3

Spatial localization: from the momentum to the coordinate representations

As mentioned, coordinate and momentum representations (or spaces) are also complementary, and can be mapped from one to the other via e.g. the exact, two-dimensional Fourier integral. In MRI, the measured (or encoded) quantity is S (kx, ky) in the momentum (or k) space and is not directly interpretable. This is in some ways similar to the FID in MRS, which needs transformation into the frequency domain. Here, however, the measurement is the Fourier integral of what is actually wanted, which is r r r the spatial distribution ( x ) of magnetization vectors = B0 .

4.3.1

Gradient magnetic fields: From k space to MR image:

Static field B0 and RF pulses cannot determine the spatial localization from which the signal has emerged. This simply means that these two fields alone are not able to provide the images. To circumvent this difficulty, Lauterbur [5] introduced the use of gradient magnetic fields to provide spatial resolution in MRI. Gradient is defined as a spatially linear variation added to the static magnetic field of strength B0 [2]. What are actually created by the magnetic gradients are time-dependent r vectors in momentum space ( k vectors). By inverse Fourier transformation, the spin density or magnetization distribution at a given spatial point is obtained [6]. The gradients are presented over short time intervals. The time interval required for a gradient to ramp up to full strength is termed rise time. The shorter the rise time (usually about 1.0 ms from 0 to 10mT/m) the better the gradient system. These gradient fields can be applied in any direction or orientation, and with three sets of gradients, a 3D image can be obtained. Gradient strength is measured in milliT/meter

40

(mT/m), and in clinical imaging systems this is usually a maximum of about 10 mT/m [7].
Slice selection The first step in MR imaging is excitation of a slice and application of a slice selection gradient, termed the z-gradient, as it is directed along the z-axis.

Next, x- and y- gradients are applied, to differentiate spins in the selected XOY plane.
Frequency encoding Frequency encoding is the next encoding step. As stated, without a gradient, all the protons experience the same field and resonate at the Larmor frequency, L. However, with a gradient, there will be a faster or slower precession, depending upon the protons position. This position-dependent frequency change is termed frequency encoding [2]. Phase encoding This is the final encoding step. The transverse magnetizations or MR signal originating from different positions along the gradient will have a positiondependent phase change. After the gradient is switched off, the nuclei will retain these different phase angles and thereby become phase encoded [2].

Proper and complete sampling of k space is vital to avoid image artefacts [2]. Moreover, mathematical advances in signal processing through the fast Pad transform have been shown to improve MR image reconstruction [6, 8 - 9]. McRobbie et al. [2] provide a musical analogy to help intuitively understand MRI in-plane localization. The reader is asked to imagine playing a multi-stringed musical instrument such as a guitar: Suppose you wish to play every possible note (or frequency) distinctly, you first have to pluck a string. This is like RF excitation. The sound it makes is like the MR signal. It is then relatively easy to play every note available on that string by running a finger up the fretboard. As one does this the length of the string is in effect shortened (this is like a gradient) and the pitch of the note (or frequency) changes. All the while the sound can be heard. This is like frequency encoding. However, there are other strings. To hear them we have to pluck one of them (perform another RF excitation) and repeat the pitch changing action. The action of changing to a different string is analogous to the operation of phase encoding in MRI(p. 125). Figure 4.2 gives an example of k space in two dimensions (kx and ky). Figure 4.3 shows the image obtained by spatial localization via inverse Fourier transformation of the k space, a sagittal view of the head of a healthy volunteer from a clinic 1.5T scanner at the Karolinska Hospital in Stockholm.

41

Figure 4.2 An example of 2-dimensional k (momentum) space with 512 grid points [8].

50

100

150

200

250

300

350

400

450

500

50

100

150

200

250

300

350

400

450

500

Figure 4.3 MR image from the k space. Sagittal view of the head of a healthy volunteer from a clinical 1.5T scanner, Karolinska Hospital, Stockholm [8].

50

100

150

200

250

300

350

400

450

500
50 100 150 200

250

300

350

400

450

500

42

4.4

A brief mathematical background

4.4.1 The Fourier transform 4.4.1.1 From the time to the frequency domain using the concept of the Fourier integral The relaxation of the nuclei after exposure to the RF pulse generates a current in the coils surrounding the scanned organ. This current is electronically converted into a time signal c (t ) FID . The measured MRS signal is a mixture of various frequencies that decay in intensity over time, with the rate of decay varying among the components. As noted earlier, this recorded signal or FID is a heavily packed, oscillating function, whose information content cannot be visually discerned directly in the time domain. The resonances are the peaks in the frequency domain that can originate from one or more exponentials (harmonics), each with a characteristic frequency (called natural, fundamental, nodal or eigen frequency). This is the dual terminology of an exponential function, which describes the measured quantity (the time signal) and the calculated result in the frequency domain. The product of the amplitude and the full width at the half maximum represents approximately the area (surface) underneath the given peak. This area or surface is also the concentration of the metabolite associated with the given peak/resonance. Heretofore, however, using in vivo MRS absolute concentrations for metabolites have been difficult to obtain, and this is due, in part, to reliance upon the conventional Fourier data analysis. As discussed, it has been standard practice to use the Fourier transform to map time signals such as the exponentially decaying MR signal, the FID, into the frequency domain. The Fourier transform is a technique used to find the frequency components of the time signal and their relative contribution to the total power of the signal. This is achieved by multiplying the signal with the exponential combination (Euler) of sine and cosine functions of various frequencies and summing up the result for all frequencies.
The Fourier integral

The spectrum F ( ) for a given time signal c(t) of finite duration is defined by the Fourier integral:

F ( ) =

1 dt c(t )ei t T 0

(4.7)

43

where e it is the complex harmonic: and is a real angular frequency.

ei t = cos( t ) + i sin( t ) , t is time,

As we presented earlier, if c(t ) were to have only a single nodal complex frequency 1 , then c(t) =d1exp(-i1t), so that F ( ) would have one peak, i.e. one Lorentzian in the limit T , where d1 is a constant complex amplitude. Recall that such F() is given by Equation (4.5), which we recapitulate here:

F ( ) = i

d1 . 1

(4.5)

The result given in Equation (4.5) is a complex-valued spectrum, whereas the power spectrum is a real quantity:

F ( ) =

d1

( 1R )2 + 12I

;
(4.8)

1R = (1 ), 1I = (1 )

whose maximum is at = 1 and this is the resonance frequency, as given by (4.5). If there are e.g. K peaks, the time domain signal is the sum of K exponentials:

c(t ) = d1e i1t + d 2 e i2t + ... + d k e i t = d k e ik t . (4.9)

k =1

This mathematical model describes the dynamics of each decaying harmonic, from k = 1 to k = K. Thus, for the model (Equation 4.9), the general formula for the power spectrum is a linear combination of K Lorentzians:

dk . F ( ) = k =1 k i
2

(4.10)

44

The magnitude spectrum is the square root of the power spectrum, i.e. F( ) . Discretization: The Discrete Fourier Integral and Transform The Fourier integral is often approximated by a sum called the Riemann sum. This is achieved by equidistantly sampling along the time axis, t, as follows:

t = tn = nt n
where t is the sampling time or the dwell time, i.e. the difference between any two adjacent time points, tn and tn+1, and where n = 0, 1, 2 N-1, comprising N time points, with N being the signal length. Then, letting

c(t ) = c (tn ) = c(nt ) cn

we obtain the Discretized Fourier Integral (DFI) from (4.7):

F ( ) =

1 N

c e .
i n n =0 n

N 1

(4.11)

Note that here, is still arbitrary. We can then substitute u-1 for ei, so that:

F ( ) =

1 N

c u
n =0 n

N 1

(4.12)

In other words, the DFI is a polynomial of degree N-1 in the variable u-1, expressed as a linear combination of power functions with constant coefficients that are the time signal points. The Discrete Fourier Transform (DFT) is associated with the restriction of an % otherwise arbitrary frequency to the Fourier grid, k ,

~ k =

2 k ; ( k = 0 ,1, 2 ,..., N 1). T

There are N Fourier grid points, i.e. as many as the sampled signal points

cn .

In other words, the DFT discretizes the frequency axis in an equidistant way:

% k = k , = 2 T .

45

% With digitization, becomes k = 2 k T = k ( 2 T ) k , where

% = 2 T min . % This min becomes the resolution in the frequency domain in the DFT for all the signals of length N, such that the DFT is finally expressed as a single ~ polynomial of degree N-1 in variable u k1 = exp( 2k N ) : % F ( k ) Fk = % uk = e
% i

1 N

% c e = N c u
in % k n =0 n n=0

N 1

N 1

n k

(4.13)

=e

2 i k N

As noted earlier, the DFT is an approximation, because whereas the exact spectrum is represented by the Fourier integral with an infinite acquisition time, the DFT uses a finite acquisition time (T < ), leading to truncation artefacts (Gibbs oscillations, aliasing, etc.) with distortion of the shape of the spectrum. Moreover, the DFT uses the Fourier grid, which has nothing to do with the characteristic frequencies k of the investigated time signal. The Fast Fourier Transform
N 1 N =1

The numerical effort of the DFT requires N2 multiplications, since we must

% multiply two vectors {cn }n =0 and {uk }k = 0 each of length N. This can be very
time consuming for large N. By selecting N to be 2m (m = 0,1,2), N2 multiplications are reduced to N log2N. This is the basis of the fast Fourier transform (FFT). In 1965 Tukey and Cooley [10] described the Nlog2N FFT, which, in fact, had been known to Gauss in 1805, Runge in 1903, and later to Danielson and Lanczos in 1942 [11].

4.4.2

How a gradient field yields spatial localization

Application of an RF pulse at the Larmor frequency perpendicular to the static magnetic field induces a time signal or FID, whose spectral representation can be mapped into the frequency domain. As discussed, this is because time (t) and frequency () are two complementary, i.e. conjugate variables, such that their functions % f (t ) and f ( ) are related by the Fourier integral. In other words, the time and frequency domains are complementary representations, with exactly the same informational content. Since coordinate and momentum representations (or spaces) are also complementary, they can be mapped from one to the other via e.g. the exact, two-dimensional Fourier integral

46

+ r r rr r % f (k ) = dreik r f (r ) =

dxdy e

ik x x + ik y y

f ( x, y ) (4.14)

% where f ( k ) is the momentum space function. The position vector

r r r = ( x, y ) and the momentum vector k = (k x , k y ) are two conjugate r variables. The momentum vector k , being a function of time, is
r defined as k = k (t ) = dt r G (t ) where is the gyromagnetic ratio,
0

r r and the vector r G (t ) is the gradient of the supplementary magnetic

field in the XOY (transverse) plane. As discussed, in MRI, the measured (or encoded) quantity is S (k x , k y ) in the momentum (or k) space and is not directly interpretable. The k vectors are created by a sequence of ramps, as the magnetic gradient is presented over time. This is in some ways similar to the FID in MRS, which needs transformation into the frequency domain. Here, however, the measurement is the Fourier integral of what is actually wanted, which is r r r the spatial distribution (r ) of magnetization vectors = B0 . The measured Fourier integral in the two-dimensional k space is:

S (k x , k y ) =

+ Lx

Lx

dxdy e
Ly

+ Ly

ik x x + ik y y

( x, y )

(4.15)

where Lx , y , are the total integration lengths along the associated coordinate axis ( x, y ), respectively. Since kx and ky are timedependent, as written above, it follows that S ( kx, ky ) = S ( kx( t ), ky( t ) ) = c(t) is also a function of time (t). In order to obtain the desired density function ( x, y ) , the inverse Fourier integral is performed. This is fairly straightforward, involving merely a change of sign in the argument of the exponential and exchange of the integrand function, as follows:

47

( x, y ) =

1 Lx

1 Lx

1 Ly

1 Ly

dk x dk y e

ik x x ik y y

S (k x , k y )

(4.16)

which yields the spatial distribution (r ) of magnetization vectors

= B.
4.4.3 Recent advances in signal processing relevant to magnetic resonance: the fast Pad transform

In a series of recent publications [4, 6, 12-14], it has been demonstrated that the fast Pad transform, as a non-linear rational approximation P/Q of the exact Maclaurin expansion of the raw time signal, can overcome many of the limitations of the FFT. In this section, we present a brief mathematical exposition of the FPT. In Part C of this book, we will focus upon the improvements that FPT offers for MRS and MRSI, and how these could impact upon clinical oncology. Recall Equation (4.7), in which the spectrum F ( ) for a given time signal c(t ) of finite duration is defined by the Fourier integral:

1 F ( ) = dt c(t )ei t T 0

(4.7)

where t is time, and exp(i t ) is the complex harmonic function, exp(i t ) = cos( t ) + i sin( t ) . Recall also, that if c(t ) were to have only a single nodal complex frequency 1 , then c(t ) = d1 exp( 1t ) , where d1 is a constant amplitude, so that F ( ) from Equation (4.5) would have one peak, i.e. one complex Lorentzian in the limit of infinitely large T:

F ( ) = i

d1 1

(4.5)

48

This result can be written equivalently, and more compactly as a ratio of the polynomial A0 ( ) and B1 ( ) of the zero and first degree, respectively:

F ( ) =

A0 ( ) , B1 ( )

(4.17)

A0 ( ) = id1 , B1 ( ) = 1
This polynomial quotient is a rational function of known in the literature as the Pad approximant (PA) and symbolized by [ 0 1]F ( ) , where 0 and 1 denote the degree of the numerator A0 ( ) and denominator B1 ( ) polynomial, respectively. The maximum in the spectrum F ( ) from Equation (4.5) is at = 1 and, therefore, 1 is identified as the complex resonant frequency. Recall that if instead of only one peak, as described by Equation (4.5), we identify K similar peaks in a Lorentzian spectrum, then the corresponding time domain signal will likewise be the sum of K exponentials with constant amplitudes {d k } (1 k K ) :

c(t ) = d1e i1t + d 2e i2t + ... + d k e i t = d k e ik t . (4.9)

k =1

Namely, inserting Equation (4.9) into Equation (4.7), and calculating the indicated Fourier integral for infinitely large T gives the mentioned general formula for the complex spectrum as a linear combination of K Lorentzians2:

F ( ) = i

dk k =1 k

(4.18)

For the previously illustrated case K = 1 , the general result (Equation (4.18)) is reduced at once to the PA [ 0 1]F ( ) from Equation (4.17). Further, if we set K = 2 in Equation (4.18), we shall obtain the spectrum:

2 In this section, we shall also discuss an extension of Equation (4.18) to a non-Lorentzian spectrum.

49

F ( ) =

A1 ( ) = i [ (d1 + d 2 ) (d1 2 + d 21 ] , B2 ( ) = 2 (1 + 2 ) + 1 2
which is the PA of the type [1 2]F ( ) . Similarly, for a general positive integer K from Equation (4.18), we shall have:

A1 ( ) , B2 ( )

(4.19)

F ( ) =

AK 1 ( ) , BK ( )

(4.20)

which is the PA of the type [ ( K 1) K ]F ( ) . Here, the polynomials

AK 1 ( ) and BK ( ) can be identified from Equation (4.18) as done previously for K = 1 and K = 2 .
Thus, we see that the sum of K damped complex exponentials with constant amplitudes {d k } for a description of c(t ) from Equation (4.9) indeed gives a linear combination of K pure Lorentzians (Equation (4.18)) which, in turn, represent explicitly the PA of the type [( K 1) K ]F ( ) as the quotient of two polynomials AK 1 ( ) and

BK ( ) of degree K 1 and K in variable , respectively.

In signal processing, the Pad approximant is renamed as the fast Pad transform [12, 13], to indicate its feature of mapping the function c(t ) from the time domain to a rational spectrum: F ( ) = AK 1 BK ( ) in the frequency domain.

The numerator and denominator polynomials of the PA or FPT can also be determined without any reference to the signal components {d k , k } by using only the measured time signal. To point at such a generality, these Pad polynomials will be denoted in the sequel by PK 1 and QK instead of the above notation AK 1 and BK , respectively. The FPT is a powerful extrapolator, due to the presence of the implicit polynomial inversion, 1 BK = BK1 in AK 1 ( ) BK ( ) from Equation (4.20). Namely, inverting a polynomial, as a finite sum, one obtains a

50

series, which is an infinite sum. This is how the FPT achieves extrapolation, since inference is gained from an infinite number of signal points by using only the available finite set {cn}( 0 n N-1). As discussed, the exact Fourier analysis requires an infinitely long time signal. If indeed such signals were available, no extrapolation would be necessary and the Fourier transform would represent the exact spectrum given by an infinite series at any frequency. However, in practice, all time signals are of finite duration T and, therefore, only approximate Fourier analysis is possible, in which case, truncation can become a serious problem, especially for short time signals. In such cases, extrapolation of finite signals is critical for extraction of the needed information. This extrapolation is equivalent to a search for a rational approximation to the infinite Fourier series. This task can be accomplished, e.g. by the FPT through its ratio of two polynomials P Q . The crux of the solution is in the implicit infinite sum contained in the inverse polynomial Q 1 since, of course, P Q = PQ 1 . From here, although both P and Q are finite sums, the inverse Q 1 is a series, i.e. an infinite sum, such that the product PQ 1 is also an infinite sum. Hence, extrapolation. Moreover, the FPT simultaneously accomplishes interpolation by computing the quotient P Q at any frequency. Besides truncation, as discussed, the FFT is restricted to the Fourier % grid, = k = 2 k T ( k = 0,1,...N 1). The FPT does not use the

% fixed Fourier mesh k in the frequency domain, and therefore, can be computed at any . As noted, the Pad theory performs interpolation and extrapolation. Resolution in the FPT is not pre-determined by T. Rather, the resolution in the FPT is the average distance ave between adjacent frequencies in a selected range. In a given frequency window, % often ave < min , such that the FPT has a better resolving power, % relative to resolution min in the FFT.

As presented earlier in this section, the spectrum in the FFT is given as % % a single polynomial F ( k ) Fk at the prescribed Fourier mesh k for all signals. In contrast, the FPT computes the spectrum

F ( z ) = n = 0 cn z n
N 1

via the unique ratio of two polynomials

F ( z ) = PK 1 ( z ) QK ( z ) at any frequency . Here, z = u 1 and u = exp(i ) , where is the sampling time ( = T N ) for a digitised signal c(t ) = c( n ) cn with 0 n N 1. The coefficients

51

of the polynomials P and Q can be computed by means of the systems of linear equations deduced from the above definition of the FPT, after equating the multipliers of the like powers of the expansion variable z , i.e. from: F ( z )QK ( z ) = PK 1 ( z ) , as per references: [1, 12]. Once these coefficients are obtained, the (non-parametric) shape spectrum can be computed by evaluating the quotient PK 1 ( z ) QK ( z ) at any selected frequency .

To extract the peak parameters, one solves the equation,

QK ( z ) = 0,

(4.21)

which is known as the characteristic equation. Every polynomial of degree n has n roots. Thus, Equation (4.21) has K roots zk (1 k K), where,

i zk = uk1 , uk = e i , k = ln(uk ).

(4.22)

Even for real cns in which case all the polynomial coefficients in PK-1 and QK are also real, there could be a number of complex frequencies { k } . In this latter case, each complex k must be accompanied by its
* complex conjugate value k .

In MRS, the cns are complex numbers, and so are the coefficients of the PK-1 and QK polynomials. Once the K roots { zk } of Qk ( z ) are found, the corresponding amplitudes following explicit expression:

{d k } are

computed from the

dk =

PK 1 ( zk ) , Q 'K ( zk )

(4.23)

where Q 'K ( z ) is the first derivative of the denominator polynomial

QK ( z ),

Q 'K ( z ) =

dQK ( z ) . dz

(4.24)

52

The parametric complex Lorentzian spectrum is obtained from the Heaviside partial fraction expansion:

PK 1 ( z ) K d k . = QK ( z ) k =1 z zk

(4.25)

The above formulae are valid for non-degenerate spectra, i.e. for all distinct roots { zk } of Equation (4.21). When some of the zk s coincide with each other (degenerate rootsleading to overlapping resonances), the above formulae should be modified. For example, if the k th root zk of QK ( z ) has Mk K multiplicity, then the Pad spectrum from Equation (4.25) is modified, via:

PK 1 ( z ) J M k d k ,mk = QK ( z ) k =1 mk =1 ( z zk ) mk

(4.26)

where M1 + M2 + + MJ = K. Here d k ,mk are the new amplitudes that generalize Equation (4.23), as:

d k ,mk =

PK 1 ( zk ) ( QKmk ) ( zk )

(4.27)

( where QKm ) ( z ) is the m th derivative of the denominator polynomial

QK ( z ) ,
( QKm ) ( z ) =

dm QK ( z ). dz m

(4.28)

However, also for the degenerate roots of Equation. (4.21), for the given N values of the cns, one always obtains the unique quotient PK 1 QK . The time signal points {cn } associated with the physical Lorentzian spectrum from Equation (4.25) with non-degenerate (distinct) roots { zk } are given by:

cn = d k e ink ,
k =1

(k) < 0.

(4.29)

53

On the other hand, the cns corresponding to the physical nonLorentzian spectrum with degenerate (multiple) roots { zk } read as:

cn = d k ,mk (n )mk 1 e ink , (k) < 0 .

k =1 mk =1

Mk

(4.30)

Convergence of those cns that contain only physical roots {uk } with (k) < 0 is guaranteed. However, if one finds spurious (unphysical) roots in Equation (4.21), and this happens for (k) > 0, then one would run into convergence difficulties (divergence), or instabilities, since this corresponds to exponentially increasing terms in the cns. These spurious roots must be regularized via, e.g. their reflection to the side of the unit circle where the genuine (physical) roots reside with the simultaneous preservation of the total energy of the signal (the socalled constrained root reflection) [12].

For reliable quantifications in MRS, it is not only the metabolite positions (k) that count, but also the peak heights d k are critical. This is simply because the k th metabolite concentration is approximately given by the product dk (k). Therefore, even for accurately determined uk ' s , the problem of obtaining the precise estimates of the d k ' s becomes extremely important. In the FPT, as seen from Equation (4.23), the k th amplitude d k depends only upon the

k th root uk . This is the reason for which the d k ' s found in the FPT
are more accurate than those computed by e.g. the parametric estimator called Linear Predictor (LP) or the Hankel-Lanczos Singular Value Decomposition (HLSVD) from MRUI software3 and the like. For the found uk ' s , the LP and HLSVD compute the d k ' s by solving the system (Equation (4.29)) of linear equations for a set of values of n. In this way, all the frequencies from the set {uk } contribute to each individual d k . Hence, any error in some of the uk ' s (stemming either from their spuriousness or insufficient accuracy in computation of otherwise physical frequencies) could severely deteriorate the d k ' s .

MRUI is the acronym for Magnetic Resonance User Interface [15]

54

As mentioned, this problem does not exist in the FPT, in which each d k is obtained separately from a single uk using the analytical formulae (Equations (4.23) and (4.27)) for a Lorentzian and non-Lorentzian spectrum, respectively. This both saves about 50% of the computational time, and, more importantly, bypasses an additional source of error from solving the system of linear equations used by the LP and HLSVD. Additionally, in contrast to the LP and HLSVD which are limited to exclusively pure Lorentzian spectra, the FPT can treat both Lorentzian (Equation (4.25)) and non-Lorentzian (Equation (4.26)) spectra on the same footing using the corresponding time signals (Equations (4.29) and (4.30), respectively), and whose constituent amplitudes are given by Equations (4.23) and (4.27), respectively. All fitting techniques use the given Fourier spectrum for extraction of peak parameters for the metabolites of interest. The FPT avoids fitting altogether, and accomplishes accurate quantification by extracting the parameters of all the physical metabolites directly from the signal or from the spectra using representations from Equations (4.29) and (4.30), for a Lorentzian and non-Lorentzian case, respectively. Specifically, the FPT first finds all the peak parameters of every physical metabolite without ever using the Fourier spectrum, or any spectrum at all, for that matter. The spectrum can be subsequently constructed for visualization purposes only, in any of the desired modes, but it is the spectral parametric quantification which matters. For example, the magnitude spectrum in the FPT is defined by the absolute value of the complex spectrum, whereas the power spectrum is the square of the magnitude spectrum: Magnitude Spectrum (FPT) =

PK 1 ( z ) , QK ( z )
2

P ( z) . Power Spectrum (FPT)= K 1 QK ( z )

The FPT can also provide the absorption and dispersion spectrum as: Absorption Spectrum (FPT) =

PK 1 ( z ) , QK ( z ) PK 1 ( z ) . QK ( z )

Dispersion Spectrum (FPT) =

55

In practice, the absorption mode should be preferred, since e.g. the magnitude mode could smooth out some important finer and weaker spectral features via interference between the terms ( PK 1 ( z ) QK ( z )) and ( PK 1 ( z ) QK ( z )) occurring in

PK 1 ( z ) QK ( z ) .

For an in-depth exposition of the FPT as it applies to MRS and MRSI, the reader is referred to [4, 6, 12-14].

References
[1] Dz. Belkic, P.A. Dando, J. Main, H.S. Taylor, Three novel high-resolution nonlinear methods for fast signal processing, J. Chem. Phys. 133, 6542-6556 (2000). [2] D.W. McRobbie, E.A. Moore, M.J. Graves, M.R. Prince, MRI from picture to proton, Cambridge University Press, Cambridge, 2003. [3] D. J. Drost, W.R. Riddle, G.D. Clarke, Proton magnetic resonance spectroscopy in the brain: Report of AAPM MR Task Group #9, Med. Phys. 29, 2177-2197 (2002). [4] Dz. Belkic, Pad-based magnetic resonance spectroscopy, J. Comp. Meth. Sci. Eng. 3, 563733 (2003). [5] P.C. Lauterbur, Image formation by induced local interactions: Examples employing nuclear magnetic resonance, Nature 242, 190-191 (1973). [6] Dz. Belkic, Fast Pad transform for magnetic resonance imaging and computerized tomography, Nucl Instr. Meth. Phys. Res. A. 471, 165-169 (2001). [7] Z-P. Liang, P.C. Lauterbur, Principles of magnetic resonance imaging: A signal processing perspective, IEEE Press, New York, 2000. [8] Dz. Belkic, K. Belkic, High resolution magnetic resonance imaging (MRI), IEEE Medical Imaging Conference (MIC), 2003, Abstract Number 1971, Portland, October 2003. [9] M.F. Callaghan, D.J. Larkman, M.C. Wiltshire, J.V. Hajnal, Edge detection at low resolution using Pad approximants, Proc. Intl. Soc. Mag. Reson. Med. 11, 754 (2004). [10] T.W. Cooley, J.W. Tukey, An algorithm for the machine calculation of complex Fourier series, Math. Computation 19, 297-301 (1965). [11] R.N. Bracewell, The Fourier Transform and its Applications, 3rd Edition, McGraw-Hill International Editions, Boston, 2000. [12] Dz. Belkic, Strikingly stable convergence of the fast Pad transform: applications to Magnetic Resonance Spectroscopy (MRS), J. Comp. Meth. Sci. Eng. 3, 299-382 (2003). [13] Dz. Belkic, Exact analytical expressions for any Lorentzian spectrum in the fast Pad transform (FPT), J. Com. Meth. Sci. Eng. 3, 109-186 (2003).

56

[14] Dz. Belkic, Quantum Mechanical Signal Processing and Spectral Analysis, Institute of Physics Publishers, Bristol, 2004. [15] A. van den Boogaart, Quantitative data analysis of in vivo MRS data sets, Magn. Reson. Chem. 35, S146-S152 (1997).

57

Chapter 5

Magnetic Resonance Spectroscopic Imaging

_______________________________________________________________________________

At this juncture, let us recall our introduction, underscoring the vital importance of molecular imaging for clinical oncology. Indeed, this is reflected in the very title of the book. Yet, thus far, we have separately discussed how anatomic images are generated through MRI and how molecular information is obtained through MRS. These were obviously necessary steps, so that we now are ready to view MRI and MRS in concert. Namely, we are now prepared to discuss molecular imaging through MRSI1. As stated by Nelson et al. [1], MRSI combines the chemical specificity of MRS with the spatial localization techniques that have been developed for MR imaging to obtain multiple MRS signals over a volume of tissue (p. 464). The value of MRSI for clinical oncology will be thoroughly explored in Part B of this book. In this chapter we will present a brief overview of how MRSI is accomplished.

5.1

Spatial resolution of spectra

As discussed, in order to determine the spatial distribution of spectra, a r third, time-varying magnetic field G(t), with a linear gradient r G (t ) is applied. In addition to the uniform (static) magnetic field, which is used in obtaining spectral information, a controlled, non-uniform

1 MRSI has also been called chemical shift imaging (CSI). In this book, we will mainly use the term MRSI.

58

gradient magnetic field provides spatial localization. This is illustrated schematically in Figure 5.1. As pointed out by Charles [2], this link between space and frequency was a critical step in the evolution of MRI and MRSI, and is based upon the use of gradient magnetic fields to provide spatial resolution in MRI, as originally introduced by Lauterbur [3].

Figure 5.1 Uniform static versus gradient magnetic field for spectral versus spatial distribution
Adapted from Charles [2] and from McRobbie et al. [4].

RF pulse at Larmor Frequency

Uniform Magnetic Field

Frequency Spectrum (Hz)

RF pulse at Larmor Frequency

Spatial Location (cm) Linear Magnetic Field Gradient

In the upper panel, two samples of the same compound in the presence of a homogeneous magnetic field yield a single peak, whereas application of a gradient magnetic field is used to obtain spatial localization of the two samples.

59

5.2

Implementation of MRSI

Recall that with single-voxel spectroscopy three orthogonal slices are selected to encompass a specific volume and often a volume of about 1.5 to 4 cc is studied. In spectroscopic imaging one does not select a specific volume, but rather, one selects a slice within which one obtains a full spectrum at each point of the selected grid, which can be of various sizes, e.g. 24x24, 16x16 etc. By processing the data one can later make images of selected peaks or ratios between peaks or one can zoom into an image and obtain the local spectra at a specific region of interest. MRSI uses phase-encoding techniques to acquire spectra from a selected slice. In many ways, this is analogous to the acquisition of 3D MR images [4], as presented in Chapter 4. Most commonly, a volume of interest is selected, and then phase encoding is applied to obtain localization for a 1D, 2D or 3D array of voxels [5]. However, in contrast to MRI, in MRSI the readout gradient is replaced by a second phase encoding lobe [6]. MRSI is most commonly performed with PRESS or STEAM. As with single voxel MRS, PRESS is now generally preferred because of its superior SNR [5]. However, rather than generating an echo from a small voxel as in MRS, with MRSI a thick slice of tissue is excited [4]. Water and lipid suppression may require special approaches, such as implementing alternative RF pulses that provide improved spatial and frequency selection. Highly spatially selective saturations bands have been used to achieve better volumetric coverage with sharper edges [5]. There are various strategies for multi-slice acquisition; scan time being a very important consideration. A major advantage of the MRSI method is acquisition of data at a very high speed enabling several slices to be obtained in a short time [4, 5]. Spielman [6] notes that there has been a misconception that MRSI is much more time consuming than MRS. He points out, however: it takes exactly the same amount of time to acquire a given SNR spectrum from a 1 cc single-voxel study as it does to acquire the same SNR spectrum from 1 cc voxels using MRSI. This insight highlights one of the major advantages of MRSI. Namely, if one is going to spend the time required for a good quality 1 cc spectrum, spectroscopic imaging allows one to simultaneously acquire the same quality spectra from a large array of 1 cc voxels (p.
28).

60

A critical requirement of MRSI is very high uniformity of the static magnetic field [7]. Variations in susceptibility should therefore be minimized in selection of volumes of interest, e.g. avoiding the sinus regions when performing MRSI of the brain, in order to facilitate shimming [5]. 5.2.1 Reconstruction and post-processing of MRSI data

The first step is to produce spectra in the k space and to reconstruct the spatial dependence of the spin density data. In order to center the data at the best spatial location, the k space array can be phase-weighted with the voxel shift. Then transformation is performed into the spatial domain, (see/recall Section 4.4.2). The array of spectra thereby obtained will have spatially dependent frequency and phase errors that must be corrected (see/recall Section 4.3.1) [5]. Advances in signal processing that can help improve the resolution of MRSI and avoid Gibbs phenomena, are discussed in Ref. [8-10].

5.2.2

An illustration of MRSI

Figure 5.2 is an example of spectroscopic imaging from our Group at the Karolinska Institute /Hospital Stockholm, Sweden, using the Matlab mathematical software. This is being further developed for full clinical use interactively at the console of the scanner while the patient is still within the scanner.

61

Figure 5.2 The encoded experimental data and theoretically obtained results from computations performed by means of the software of our group (Dz. Belkic, K. Belkic, B. Nordell, L. Douglas).
This figure was kindly provided by Drs. L. Douglas and B. Nordell, as implemented at the Department of Radiology, Astrid Lindgren Childrens Hospital, Stockholm, Sweden. From Ref. [11], with permission.

Anatomical image

Spectroscopic grid

Integrated Spectra NAA Choline

Local MR Spectra
In the upper right panel entitled Spectra, two narrow spectral regions are selected around 2 ppm and 3.2 ppm, near NAA and choline, respectively, as indicated by the two sets of vertical lines. The concentrations of metabolites from these narrow strips near NAA and choline are further depicted as local MR spectra on the lower left panel through the indicated field of view.

References
[1] S.J. Nelson, E. Graves, A. Pirzkall, et al., In vivo molecular imaging for planning radiation therapy of gliomas: an application of 1H MRSI, J. Magn. Reson. Imaging 16, 464-476 (2002). [2] H. C. Charles, The whole body NMR Spectroscopy examination, Section for Magnetic Resonance Technologists Educational Seminars 3, 5-18 (2000). [3] P.C. Lauterbur, Image formation by induced local interactions: Examples employing nuclear magnetic resonance, Nature 242, 190-191 (1973). [4] D.W. McRobbie, E.A. Moore, M.J. Graves, M.R. Prince, MRI from picture to proton, Cambridge University Press, Cambridge, 2003.

62

[5] S.J. Nelson, Multivoxel magnetic resonance spectroscopy of brain tumors, Mol. Cancer Ther. 2, 497-507 (2003). [6] D. Spielman, In vivo proton MR spectroscopy: basic principles and clinical applications, Section for Magnetic Resonance Technologists Educational Seminars 3, 19-37 (2000). [7] H-M. Baik, B-Y. Choe, B-C Son, et al., Feasibility of proton chemical shift imaging with a stereotactic headframe, Magn. Reson. Imaging 21, 55-59 (2003). [8] Dz. Belkic, Fast Pad transform for magnetic resonance imaging and computerized tomography, Nucl. Instr. Meth. Phys. Res. A 471, 165-169 (2001). [9] Dz. Belkic, K. Belkic, High resolution magnetic resonance imaging (MRI), IEEE Medical Imaging Conference (MIC), 2003, Abstract Number 1971, Portland, October 2003. [10] M.F. Callaghan, D.J. Larkman, M.C. Wiltshire, J.V. Hajnal, Edge detection at low resolution using Pad approximants, Proc. Intl. Soc. Mag. Reson. Med. 11, 754 (2004). [11] K. Belkic, Magnetic resonance spectroscopy and spectroscopic imaging: a review of basic principles and achievements in oncology, J. Comp. Meth. Sci. Engin. 3, 505-533 (2003).

63

Chapter 6

Safety Considerations in Magnetic Resonance

_______________________________________________________________________________

6.1

Effects of exposure to electromagnetic fields and other risks related to magnetic resonance

This chapter provides a very brief overview of hazards and safety considerations related to magnetic resonance. The reader is referred to specialized materials, some of which are cited herein, for more detailed and comprehensive information on this topic. Potential biological effects of exposure to magnetic fields The potential biological effects of MRS and MRSI are related to exposure to static, gradient and radiofrequency electromagnetic fields. Each of these can represent a possible health risk at high levels of exposure, although there is much uncertainty and controversy about the extent to which low-level exposures are harmful [1]. It has been argued that there is a small value of magnetic susceptibility of human tissues. However, there have been no replicated studies demonstrating a health hazard associated with magnetic field exposure, nor with cumulative exposure to these fields [2]. While the patient is exposed to a static magnetic field, time-varying gradient fields and RF fields, the staff working in MR units is exposed only to the fringe field, i.e. the field extending beyond the scanner [3]. Effects of magnetic fields on ferromagnetic objects The strong static magnetic field causes ferromagnetic attraction such that ferromagnetic materials can become strongly magnetized [4]. When an object containing iron or steel comes close to the magnet, this can become a projectile. If the object is large this can represent a lethal force [3].

64

Even a ferromagnetic pillow, which is not detectable by a hand-held magnet, could be harmful especially for patients with cervical instability [5].
Effects of magnetic fields on biomedical implants

There may be hazards for patients with certain biomedical implants and devices, especially those related to the cardiovascular system. Some of these represent contraindications for the use of MR (for further details, see Ref. [6]). Magnetic fields can cause implantable pulse generators (pacemakers) to malfunction and render the electrode dangerous to the patient [7]. Due to the Lenz effect, there is a possibility for deleterious consequences upon metallic heart valves [8], although an in vitro study found no significant force on heart valve prostheses or biologically important increase in temperature [9]. Aneurysm clips can be moved. There has been a death reported in a patient with aneurysm clips that ruptured the blood vessel [3].

Neural stimulation effects Peripheral nerve stimulation from rapidly switched gradients and auditory noise levels [10] can occur. Recently, Mansfield et al. [11] described advances in gradient coil design to significantly reduce acoustic noise and induced currents, enabling faster scanning at higher magnetic field strengths, while at the same time minimizing neural stimulation effects. Thermal injuries Some thermal injuries have been reported [12]. There is a small risk of burns related to the coupling of RF energy into wires or cables, such as those used for physiological monitoring. Non-ferromagnetic metallic implants can result in greater heating [3]. Decoupling procedures used in 31P MRS entail higher power levels and therefore require particular attention to the specific absorption rate of the tissue under study, to keep local heating below the recommended maximum [13]. Leakage of liquid nitrogen or helium Work with cryogens, liquid nitrogen or helium, can lead to cold burns or hypothermia. Inhalation of cold gas can provoke asthma in susceptible individuals [3]. Leakage of liquid nitrogen leading to oxygen displacement has occurred during installation of an MRI system [14]. Claustrophobia and other anxiety responses MRI can elicit anxiety responses including claustrophobia in a considerable number of patients.

65

6.2

Precautions in magnetic resonance

Earplugs to protect against high acoustic noise These are recommended for all patients and personnel in the scanning room [3]. Physiologic monitoring This requires special care when applied in the scanner. In particular, cables should not form loops. Dry flame-retardant pads must be placed between the cables and the patient. The cables should be specifically compatible with MR systems with note taken of the maximum operating proximity). Cables should always be inspected for intact insulation [3]. Contrast agents Adverse reactions to contrast agents are very unusual. However, these may be restricted during pregnancy and for nursing mothers. Patient groups in whom particular caution is recommended [3]
Newborns and infants Pregnant women Compromised thermoregulatory systems Besides newborns and low-birth weight infants, this can include some patients with cancer, some hypothalamic disorders, inter alia. Patients who may have had ferromagnetic objects accidentally lodged in them This might include, e.g. metalworkers or soldiers.

6.3

Contraindications to magnetic resonance

Contraindications to MRI have been cited to include:

Implanted cardiac pacemakers, cardioverter defibrillators,1 Certain other ferromagnetic or metallic implants2, [17]

1 However, there have been reports of individual patients with pacemakers undergoing MR examinations without sequelae [15, 16]. 2 It appears that there are no adverse side effects of MRI with respect to metallic middle ear implants [18]. However, in vitro experiments at a 1.5T MR scanner showed that the torque on the internal magnet is hazardous for patients with cochlear implants [19].

66

Severe claustrophobia. (Although this usually can be handled with mild sedation.) 1st 12 weeks of gestation (According to some authors [20]).

More detailed, extensive lists of contraindications are given in Ref. [21], and references cited therein. Young [22] notes that there are numerous inconsistencies among various sets of regulatory requirements, underscoring the need for a common standard for the international regulatory system with respect to MRI.

References
[1] G.J. Beers, J.L. Phillips, F.S. Prato, I. Nair, Biological effects of low-level electromagnetic fields: current issues and controversies, Magn. Reson. Imaging Clin. N. Am. 6, 749-774 (1998). [2] J. F. Schenck, Safety of strong, static magnetic fields, J. Magn. Reson. Imaging 12, 2-19 (2000). [3] D.W. McRobbie, E.A. Moore, M.J. Graves, M.R. Prince, MRI from picture to proton, Cambridge University Press, Cambridge, 2003. [4] The safe use of equipment in the magnetic resonance environment, Health Devices, 30 421-444 (2001). [5] B. Condon, D.M. Hadley, R. Hodgson, The ferromagnetic pillow: a potential MR hazard not detectable by a hand-held magnet, Br. J. Radiol. 74, 847-851 (2001). [6] S. Ahmed, F.G. Shellock, Magnetic resonance imaging safety: implications for cardiovascular patients, J. Cardiovasc. Magn. Reson. 3, 171-182 (2001). [7] D.S. Bhachu, E. Kanal, Implantable pulse generators (pacemakers) and electrodes: safety in the magnetic resonance imaging scanner environment, J. Magn. Reson. Imaging 12, 201-204 (2000). [8] B. Condon, D.M. Hadley, Potential MR hazard to patients with metallic heart vales: the Lenz effect, J. Magn. Reson. Imaging 12, 171-176 (2000). [9] D. Pruefer, P. Kalden, W. Schreiber, et al., In vitro investigation of prosthetic heart valves in magnetic resonance imaging: evaluation of potential hazards, J. Heart Valve Dis. 10, 410-414 (2001). [10] R.R. Price, The AAPM/RSNA physics tutorial for residents: considerations, Radiographics 19, 1641-1651 (1999). MR imaging safety

[11] P. Mansfield, R.M. Bowley, B. Haywood, Controlled E-field gradient coils, MAGMA 16, 113120 (2003). [12] M.F. Dempsey, B. Condon, Thermal injuries associate with MRI, Clin. Radiol. 56, 457-465 (2001).

67

[13] J.R. Griffiths, A.R. Tate, F.A. Howe, M. Stubbs, as part of the Multi-Institutional Group on MRS Application to Cancer, Magnetic resonance spectroscopy of cancer - practicalities of multicentre trials and early results in non-Hodgkins lymphoma, Eur. J. Cancer 38, 2085-2093 (2002). [14] J.R. Gill, S.F. Ely, Z. Hua, Environmental gas displacement: three accidental deaths in the workplace, Am. J. Forensic Med. Pathol. 23, 26-30 (2002). [15] I. Garcia-Bolao, V. Albaladejo, A. Benito, et al., Magnetic resonance imaging in a patient with a dual-chamber pacemaker, Acta Cardiol. 53, 33-35 (1998). [16] A. Geppert, F. Rauscha, Pacemaker dysfunction in clinical practice, Wien Klin. Wochenschr. 113, 15-26 (2001). [17] R.H. Goh, S. Somers, E. Juriaans, J. Yu, Magnetic resonance imaging. Application to family practice, Can. Fam. Physician 45, 2118-2228, 2131-2132 (1999). [18] P. Kwok, A. Waldeck, J. Strutz, How do metallic middle ear implants behave in the MRI? Larynorhinootologie 82, 13-18 (2003). [19] C. Teissl, C. Kremser, E.S. Hochmair, I.J. Hochmair-Dosoyer, Magnetic resonance imaging and cochlear implants: compatibility and safety aspects, J. Magn. Reson. Imaging 9, 26-38 (1999). [20] K. Togashi, MR imaging of the ovaries: normal appearance and benign disease, Radiol. Clin. N. Am. 41, 799-811 (2003). [21] W. P. Dillon, Neuroimaging in neurologic disorders, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 2337-2342. [22] I.R. Young, Notes on current safety issues in MRI, NMR Biomed. 13, 109-115 (2000).

69

Part B MRS and MRSI -- Current State of the Art in Clinical Oncology Chapter 7

Magnetic Resonance in Cancer Diagnostics: Generalities

_______________________________________________________________________________

The real strides made by in vivo MRS and MRSI in clinical oncology are, in fact, highly tumor- or site specific. The most dramatic achievements have been made with respect to brain tumors, for which this molecular imaging modality has become an indispensable diagnostic component. MRS and MRSI are also now becoming very important in various aspects of prostate cancer diagnostics. For other malignancies, in vivo MRS and MRSI while showing promise, still have had relatively limited applications. On the other hand, MRI is widely used in staging many, if not most, solid cancers. Moreover, for some special indications, notably young women at high risk for breast cancer, MRI is being explored as a possible first-line diagnostic modality. In the subsequent chapters of Part B of this book, we will provide an in-depth review of the state-of the art of molecular imaging through MR in clinical oncology, on a tumor- or site-specific basis. In this chapter, we first present some generalities concerning MR in cancer diagnostics.

7.1

Relaxation rates of neoplastic tissues

A seminal finding that has facilitated the identification of malignancy by magnetic resonance is that neoplastic tissue generally relaxes at rates different from the surrounding tissue [1].

70

An early NMR study [2] in rats revealed that T1 relaxation times were nearly three times longer in 3 Novikoff hepatomas compared to 5 normal liver specimens. The T2 relaxation times were about twice as long in the tumors. Subsequently, these authors reported that a combined T1 and T2 normalized NMR malignancy index distinguished cancerous and normal tissues in 36 colon samples, 22 of 23 breast samples and 26 of 29 lung samples [3]. However, in a more recent study [4] human surgical samples of malignant and normal large bowel, while showing statistically significant differences in T1 and T2 relaxation times, these differences were considered too small to use the NMR method alone for diagnosis (p. 51). Many cancers appear hyper-lucent on T2 weighted images. However, if the surrounding tissue, e.g. endometrium is hyper-intense on T2 weighted images, neoplastic tissue, i.e. endometrial carcinoma, can be of intermediate intensity [1]. Prostate cancer, on the other hand, is generally of low signal intensity on T2 weighted images. There are also exceptions based upon histological type. Sometimes, tumors may be iso-intense to their surroundings. On the other hand, a difference in signal intensity compared to surrounding tissue is by no means pathognomonic for neoplasia.

7.2

Contrast Enhancement

Contrast agents are frequently used to improve detection of tumors. Gadolinium, the most commonly used contrast agent, is a paramagnetic metal, which reduces the relaxivities of the tissues where it is taken up. As a consequence, tissues into which gadolinium diffuses will appear Since tumors are frequently bright on T1 weighted images. hypervascular, they will often preferentially take up the contrast agent compared to the surrounding tissue. Moreover, the tumor vasculature tends to be disorganized and leaky, and as a consequence, tumors will often have increased extracellular fluid. For these reasons, cancerous tissue frequently, but not always, shows contrast enhancement. The dynamics of contrast uptake can help distinguish tumor from its surroundings [1]. Contrast enhancement is not a unique property of cancer; it is also seen in many other pathologic processes, as will be discussed in the next chapters.

7.3 Informative metabolites and metabolite ratios from in vivo MRS for cancer diagnostics
In this section we will only discuss those metabolites or ratios for which elevated levels are associated with several types of cancers, on

71

the basis of in vivo clinical MRS studies. presented in the chapters on specific cancers. 7.3.1
1

The in vitro data are

H MRS

Individual metabolites
Choline [3.2 ppm] Choline is currently the most important indicator of malignancy in 1H MRS. It reflects phospholipid metabolism of cell membranes, and is a marker for membrane damage, cellular proliferation and density. Total choline represents the sum of phosphorylcholine + glycero-phosphorylcholine + 5% free choline. Elevated choline is not unique to cancers, however, and can be seen with a wide variety of other pathologies, as well. The entire signal contributes to the single narrow peak, and there is no J coupling, so that pathological changes are amplified almost 10-fold [5]. Lactate [1.3 ppm] Lactate is the end product of failed oxidative metabolism, and is a sign of increased anaerobic glycolysis. It is well known that tumor cells frequently rely upon energy generated from glycolysis to fuel their requirements for rapid replication. This is because cancerous degeneration of tumor cells is associated with loss of the more efficient aerobic mode of ATP production via the Krebs cycle [6]. However, tumors with low metabolic activity may not show elevated lactate. On the other hand, lactate is elevated in ischemia and hypoxia, as well as in inflammatory and infectious processes, inter alia. Lactate is usually, but not always, very low or absent in healthy tissue. At intermediate TE ( 140 ms), lactate inverts below baseline and appears as a doublet due to J coupling. At short TE, lactate overlaps with lipids. Lipids [1.3 ppm] Lipids at 1.3 ppm are usually pathological. In a number of tissues the appearance of increased levels of lipids at 1.3 ppm suggests necrosis, which may be cancer-related, but also is typical of infection, inflammation, as well as ischemia and hypoxia. Lipids at 1.3 ppm are detected at short TE, where they overlap with lactate, as noted.

Metabolite ratio
Choline to creatine ratio The choline to creatine ratio has been found to be an indicator of various malignancies. Creatine is used in the brain and a number of other tissues as an internal reference. This ratio can also be elevated in inflammatory and ischemic processes.

72

7.3.2

31

P MRS

Individual metabolites
Phosphomonoesters (PME)[ 7.0 8.0 ppm] The main finding in a wide variety of human cancers is a fall in the PME peak in the early stages of successful therapy, regardless of modality [7]. PMEs are mainly comprised of phosphorylcholine and phosphorylethanolamine. These are metabolic components of both the anabolic and catabolic branches of phospholipid metabolism, and of proliferating tissues, such that tumors (both experimental and naturally occurring) have increased levels of PME [8]. Phosphorylcholine appears to be quite specific to cells that express oncogenes, and it may be a key step in stimulating growth factors [9]. On the other hand, phosphoethanolamine shows an inverse association with cell growth, and mayreflect intrinsic regulatory factors [8] (p. 1397).

Metabolite ratio
Phosphomonoesters to -ATP The concentration of ATP is used as an indicator of oxidative metabolism. A decline in the ratio of PME to -ATP has been found to be an indicator of successful therapy in a number of cancers. A fall in this ratio frequently precedes a diminution in tumor size [7].

7.4 MRSIImportance of full volumetric coverage

The advantages of MRSI as compared to MRS for tumor diagnostics cannot be overemphasized. As will be seen in the next chapters, the major breakthroughs in brain and prostate cancer diagnostics have relied heavily upon assessment of the metabolic characteristics of larger areas of suspected tumor, peritumoral regions and normal tissue. This is provided by MRSI. Spielman [10] states: the spatial resolution of in vivo {MR} spectroscopy is simply too coarse to avoid false negative findings resulting from small lesions partially volumed with normal tissues(p. 30) and emphasizes the need for volumetric coverage using proton MRSI for central nervous system (CNS) tumor diagnostics. Complete volumetric coverage of a suspected lesion is critical for accurate diagnosis, and Spielman notes that metabolically abnormal regions may not precisely correspond to much larger T2 abnormalities. Thus, he concluded: sampling the most metabolically abnormal tissue may be a better choice than just sampling the tissue with the highest level of contrast-enhancement (p. 31). Vigneron et al. [11] similarly point out the weakness of single voxel techniques that provide one spectrum {as in MRS} which is presumed to be representative of the entire tumor. They note that this technique

73

provides no information on the extent of the metabolic abnormality and suffers both from the inaccuracy of conventional MRI to define the exact extent of solid neoplasms and the considerable tissue heterogeneity within the tumor mass (pp. 98-99). Thus, these authors underscore the need for 3D spectroscopic imaging for identification of viable tumor on the basis of both morphologic and metabolic parameters for targeting and following local therapy (p. 100).

References
[1] A.E. Li, D.A. Bluemke, Magnetic Resonance Imaging in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 669-679. [2] R. Damadian, Tumor detection by nuclear magnetic resonance, Science 171, 1151-1153 (1971). [3] J.A. Koutcher, M. Goldsmith, R. Damadian, NMR in cancer. X. A malignancy index to discriminate normal and cancerous tissue, Cancer 41 174-182 (1978). [4] P. Kowalski, P. Skupin, J. Sowier, K.J. Olszewski, NMR relaxation time studies of large bowel neoplasms, Physiol. Chem. Phys. Med. NMR 29, 51-54 (1997). [5] E. R. Danielsen, B. Ross, Magnetic Resonance Spectroscopy Diagnosis of Neurological Diseases, Marcel Dekker, Inc., New York, 1999. [6] J. Czernin, M.E. Phelps, Positron emission tomography scanning: Current and future applications. Annu. Rev. Med. 53, 89-112 (2002). [7] W. Negendank, Studies of human tumors by MRS: A review, NMR Biomed. 5, 303-324 (1992). [8] N.R. Aiken, R.J. Gillies, Phosphomonoester metabolism as a function of cell proliferative status and exogenous precursors, Anticancer Res. 16, 1393-1397 (1996). [9] J.D. Bell, KK. Bhakoo, Metabolic changes underlying hepatic tumors, NMR Biomed. 11, 354-359 (1998).
31

P MR spectral alterations in human

[10] D. Spielman, In vivo proton MR spectroscopy: basic principles and clinical applications, Section for Magnetic Resonance Technologists Educational Seminars 3, 19-37 (2000). [11] D. Vigneron, A. Bollen, M. McDermott, et al., Three-dimensional magnetic resonance spectroscopic imaging of histologically confirmed brain tumors, Magn. Reson. Imaging 19, 89-101 (2001).

75

Chapter 8

Brain Tumors
_______________________________________________________________________________

We begin our consideration of specific cancers by a review of brain tumors. Obviously, it is here that the most delicate of clinical decisions are made, and these require maximal information of the highest possible reliability. In no other area of oncology have MRS and MRSI become so widely incorporated into clinical practice. Non-invasive molecular imaging via MRS and MRSI now represents the key modality for nearly all aspects of brain tumor diagnostics. Before reviewing these important advances, we present a very brief review of epidemiologic and clinical aspects of brain tumors.

8.1

Overview of epidemiological & clinical aspects

Incidence and prevalence/morbidity and mortality In comparison to other types of malignancies, primary brain tumors are relatively rare in adults. Nevertheless, they draw a great deal of attention because of the fear associated with the location, the young age at which these can occur, and the overall poor prognosis. There are approximately 18 000 newly diagnosed brain tumors annually in the U.S. As noted by Savitz and Trichopoulos [1], the epidemiology of brain tumors is challenging to summarize, even at the simple descriptive level {such that} comparing incidence rates across settings, populations or time periods in which the diagnostic methods are unequally available is problematic (p. 486). A relative peak in the incidence of brain tumors is seen among very young children. Brain tumors are much more infrequent among

76

adolescents and young adults. Thereafter, the incidence rises in relation to age reaching a plateau at age 65-79 years. Tumors of the CNS are the most common solid neoplasms to occur in children, and the second leading cause of cancer-related deaths among children under age 15 [2]. The overall mortality rate from malignant brain tumors is about 5 per 100 000 per year. A similar number of benign tumors of the CNS occur, with a much lower mortality rate. Nevertheless, due to their anatomic location, benign brain tumors represent a major threat to the patients health. Metastatic brain lesions are more common than primary brain tumors. Approximately 15% of patients who die of cancer have symptomatic brain metastases [3]. Etiology/risk factors Inherited susceptibility A positive family history of brain tumors is associated with a modest increased relative risk; reports range from a maximum of 2, but usually below 1.6 or even approaching 1 [1]. A number of hereditary syndromes are associated with a risk of developing brain tumors. These include:
Li-Fraumeni Syndrome This is a familial cancer syndrome due to germ-line abnormalities in p53. Neurofibromatosis Types I and II Retinoblastoma Multiple endocrine neoplasias 1 (Wermer

Syndrome)

Turcot Syndrome Gorlin (basal cell nevus) Syndrome.

However, known genes account for less than 1% of primary brain cancers and familial syndromes for less than 4%. Environmental / Occupational exposures
Ionizing radiation The best documented environmental risk factor for brain tumors. Although airline pilots and cabin crew are exposed to low energy ionizing radiation from cosmic rays, and are found to be at increased risk for cancers such as malignant melanoma, these occupations have not been reported to be at risk for brain tumors [4-6].

77

Vinyl chloride Consistent data link occupational exposure to vinyl chloride to increased risk of brain tumors [1]. Occupations at high risk There are reports of an elevated risk among petrochemical, agricultural and electrical workers, as well as among health professionals [1]. Cellular telephones: very limited evidence The use of cellular telephones has received much attention in relation to brain tumors, motivated by the public health implications, case reports, and a suggested biological mechanism of the emission of low-power electromagnetic radiation in the radio-frequency range, which could lead to a slight increase in temperature of exposed tissue (site specific) [1]. As yet, however, there is only very limited epidemiological evidence from investigations supporting such an association [7, 8]. Hardell et al. [8] reported an odds-ratio (OR) of 1.59 (95% Confidence Intervals (CI) = 1.05 2.41) for digital mobile phones and all malignant brain tumors in their case-control study from Sweden. On the other hand, Inskip et al. [9] in their case-control study from the U.S. reported an OR = 1.0 (95% CI 0.6 1.5) for cumulative cellular telephone use of more than 100 hours and all types of brain tumors. These authors conclude that the data do not support the hypothesis that the recent use of hand-held cellular telephones causes brain tumors, but they are not sufficient to evaluate the risks among long-term, heavy users and for potential induction periods (p. 79). It has been also suggested that there may be a publication bias towards null studies in this area [7]. Ingestion of cured meats: possible risk There are some positive prospective data including dose-response relationships with respect to adult gliomas, but null findings have also been reported from prospective studies, as reviewed in a meta-analysis by Huncharek et al. [10].

Medical conditions and related exposures

Hormone replacement therapy and meningiomas Recent prospective data from the Nurses Health Study with over 1 million person-years of follow-up reveal a significant risk among registered nurses of meningioma associated with hormone replacement therapy (relative risk = 1.86 95% CI = 1.07 3.24, adjusted for age and body mass index) [11]. Meningiomas have been found to be positively associated with breast cancer, which may also be hormonally related [1]. Acromegaly Associated with an increased risk of brain cancer, as well as at other sites, in a cohort study from Sweden and Denmark. This may be related to pituitary irradiation as well as to increased levels of insulin-like growth factor, which is linked to increased anti-apoptotic and proliferative activity [12].

78

Clinical Presentation Brain tumors typically present with one of three clinical patterns:
Focal neurological deficit This is usually with a sub-acute progression, and is due to compression of neurons and white matter by the expanding tumor and associated brain edema, and hemorrhage, in some cases. Seizures These can be caused by disruption of cortical circuits, and mostly occur with cortical, as opposed to sub-cortical, tumors. Non-focal neurological dysfunction Included here are headaches, usually exacerbated by recumbancy, coughing, sneezing or straining, and related to increased intra-cranial pressure. Subtle deficits, including personality changes, dementia or mood disorders are often seen with tumors of the frontal lobe. Occasionally, brain tumors are asymptomatic, being discovered during imaging for an unrelated problem [1, 3].

Classification and Grading Tumors are generally classified by their cell of origin. A large number of phenotypically distinct cell types are capable of transformation into tumors. There have been substantial changes made in the classification systems, which, as noted, create difficulties when attempting to compare data, particularly from registries. Primary Brain Tumors from Glial Precursors
Astrocytomas Derived from astrocytes, the most common type of primary brain neoplasms. These are most often graded by the World Health Organization (WHO) system: Grade I, excellent prognosis after surgical removal, includes juvenile pilocytic astrocytomas (seen mainly in children), subependymal giant cell astrocytoma and pleomorphic xanthoastrocytoma. Grades II and III, intermediate grades, include anaplastic astrocytomas (Grade III). Grade IV, glioblastoma multiforme (GBM), highly aggressive tumors. Higher-grade astrocytomas show hyper-cellularity, atypical nuclei and cytoplasm, necrosis and endothelial proliferation, the latter being a particularly strong predictor of aggressiveness. Grading is limited by the fact that astrocytic tumors show a great deal of heterogeneity from region to region. They should be classified by their most

79

malignant features, a point which will be important to recall later in this chapter. Moreover, astrocytic tumors vary over time, typically progressing to a higher grade with increased aggressivity. High-grade astrocytomas most often have ill-defined margins, with cancerous cells migrating along white matter pathways to adjacent brain regions. Gliomatosis cerebri is a rare type of astrocytoma with diffuse infiltration of the brain by malignant astrocytes. Oligodendrogliomas Usually more benign than astrocytomas, arise from oligodendrocytes. Oligodendrogliomas occur mainly in the supra-tentorial regions and often contain calcifications. Mixed astrocytic and oligodendral tumors also occur (called mixed gliomas); the greater the predominance of oligodendral elements the more benign the tumor. Nuclear atypia, mitoses and necrosis are associated with a more aggressive tumor, and if prominent these are called anaplastic oligodendrogliomas.

Meningiomas Account for approximately 25% of all primary CNS tumors, and are derived from mesodermal elements, probably arachnoid fibroblasts. Meningiomas are generally benign and are attached to the dura. They may invade the skull, but rarely the brain. They often have a characteristic appearance on MRI (see Section 8.2). Meningiomas occur more frequently among women than men, with a peak incidence in middle age. Primary CNS Lymphomas Primary lymphomas of the CNS are generally of intermediate to high grade, presenting within the neuro-axis without manifestations of systemic lymphoma. These most commonly occur in immunocompromised patients, and are often multi-centric. Metastatic Brain Tumors About 20% of all detected brain tumors are metastases [13]. The anatomic distribution of metastatic brain tumors usually parallels regional cerebral blood flow, with common occurrence at the grey matter-white matter interface, and at the border between the middle and posterior cerebral artery distributions. Lung cancer is the most frequent origin of brain metastases, followed by breast cancer and then melanoma. In about one-third of patients with metastatic brain tumors a known primary cancer has not been found. References [1-3] can be consulted for more details, especially about the less common types of CNS tumors.

80

General Approach to Diagnosis and Differential Diagnosis After appropriate clinical evaluation of the patient, imaging with MRI and CT has become the hallmark for diagnosing a large number of neurological disorders. MRI is preferred for the detection of brain tumors for reasons that will be discussed in section 8.2. Actual tumor diagnosis and grading, however, has depended upon histopathologic analysis of the surgical material [14]. Increasingly, MRS and MRSI are being incorporated into various aspects of tumor diagnostics (including differential diagnostics), as will be thoroughly discussed in Sections 8.3 8.8. Focal neurological deficit Rapid onset of focal neurological deficit is most typical for stroke, while, as noted above, focal neurological deficit with a sub-acute progression is the more common presentation due to brain tumor and surrounding edema, as well as hemorrhage, which in some cases compresses adjacent structures. Although until recently, CT was considered necessary to rule out hyper-acute intra-cerebral hemorrhage (within the first few hours), MRI is now emerging as the method of choice of early evaluation of stroke [15]. New onset of seizures in adults With increasing age, brain tumors become progressively higher on the differential diagnosis of new onset seizures. Among adults over age 35, cerebrovascular accidents are the most common cause of seizures. Alcohol withdrawal is also high on the differential diagnosis of seizures among young and older adults. Metabolic disorders may cause seizures at any age. While electroencephalography is always required to evaluate patients with possible seizure disorders, brain imaging with MRI is also part of the work-up to rule out an underlying structural abnormality [16]. MRI is also used for the lateralization of the epileptogenic focus [17]. Headache Brain tumor enters into the long list of differential diagnosis of headache, being the chief complaint for about 30% of patients. The headache pain associated with brain tumors is usually intermittent and moderate to severe in intensity, often of a pounding character. It may be exacerbated by motion, and is sometimes accompanied by nausea and vomiting. The latter is also typical of migraine headaches, which, of course, are much more common. However, new migraine headache with progressively increasing severity raises the suspicion of brain tumor. Headache arising de novo in patients with known malignancy

81

suggests the possibility of brain metastases. Other serious underlying causes of headache include cerebral hemorrhage or ischemia, temporal arteritis or other vascular disorders, intracranial infection and meningitis, glaucoma, arterial hypertension, inflammatory disorders, inter alia [18]. Dementia Neoplasia is a relatively uncommon cause of dementia. Neuro-imaging with MRI and MRS plays an increasingly important role in the work-up of dementia, in particular for distinguishing the most common cause, Alzheimers disease, from other etiologies. Diagnosis of brain tumor manifested clinically only by subtle changes in personality and mood disorders remains a major challenge. Neuro-imaging with MRI and MRS sometimes shows abnormalities with primary psychiatric disorders [17].

Treatment and Prognosis

Complete surgical resection is the aim for successful treatment of brain tumors. Realization of this goal, however, depends very much upon the clinical and histopathologic features of the tumor. Oligodendrogliomas mainly composed of oligoastrocytes and without aggressive features are often amenable to total surgical removal. These tumors are frequently responsive to chemotherapy. Radiation therapy (RT) may be also applied for residual tumor. Overall, five-year survival for oligodendrogliomas is reported to be over 50%. Treatment of low-grade astrocytomas involves surgical resection and RT. Regimens vary considerably among centers, and strategies also differ greatly in relation to the clinical features. Since high-grade astrocytomas only rarely have clearly defined margins, total surgical resection is not possible in most cases. Partial resection to control mass effect is often performed, as well as RT, chemotherapy and glucocorticoids. Overall survival is poor, generally below 1 year. Total surgical resection represents a curative treatment for meningiomas. If the resection is sub-total, local RT is usually given and reduces recurrence rates to fewer than 10%. If the meningioma is not surgically accessible, targeted radio surgery with the gamma knife or heavy particle radiation may be considered. Primary CNS lymphomas have a poor prognosis, particularly among patients with acquired immunodeficiency disorder (AIDS). Radiation therapy, chemotherapy and glucocorticoids have been attempted.

82

Metastatic tumors to the brain are generally treated palliatively with the aim of ameliorating suffering and disability and, if possible, to prolong life [3]. However, it has been argued that selected patients with multiple brain metastases may benefit from aggressive treatment of these tumors [19].

8.2 Magnetic resonance imaging in brain tumor diagnostics

As stated earlier in this chapter, MRI is now preferred over CT for the detection of brain tumors. The reasons include multi-planar capabilities, which improve identification of the origin of the tumor and extension to adjacent structures. Tissue contrast is also better with MRI, as well as imaging detail in the posterior fossa, and visualization of secondary effects such as hydrocephalus, edema and hemorrhage. However, detection of calcifications and tumor invasion into bone is better achieved with CT. Molecular imaging with FDG PET has been applied in brain tumors. Its sensitivity is fairly high (for detecting recurrence of brain tumor on the average 80%, with a range from 73 to 86%), but specificity is very low (average 39% with a range from 22 to 56%) [20]. 8.2.1 Characteristics using T2 weighted images

The usual appearance of brain tumors on MRI is hyper-intensity on T2 weighted images [14]. Edema, which occurs especially with rapidly growing tumors, also appears hyper-lucent with T2 weighting [2]. T2 hyper-intensity is not unique, however to brain tumors. A T2 weighted signal increase is also seen in many other CNS abnormalities, including ischemic and hyper-acute as well as sub-acute hemorrhagic stroke, infection, inflammation and some degenerative brain disorders, inter alia [13, 17, 21]. Among children, the clinical and MRI patterns of bilateral thalamic astrocytomas are very similar to those of neurometabolic disorders and of encephalitis [22]. Not all brain tumors appear T2 hyper-intense. Meningiomas, for example, are often iso-intense to the surrounding brain tissue, when viewed on MRI without contrast [14]. Small brain metastases may not appear on T2 weighted imaging [13]. CNS lymphomas may be hypointense on T2 weighted MRI [17]. Moreover, the peri-tumoral region often appears unchanged on standard MRI, but represents an uncertain zone where tumor may be present in patients with cerebral glioma [23].

83

8.2.2

Contrast Enhancement (CE)

Gadolinium contrast is used in all MRI work-ups of brain tumors. The normal blood brain barrier (BBB) prevents uptake of contrast into the brain. The BBB is often broken down with brain tumors, and there is also frequently new blood vessel formation, such that the majority of brain tumors show contrast enhancement [14]. However, since disruption of the BBB is not unique to brain tumors, contrast enhancement on MRI is not pathognomonic for CNS malignancy. Cerebral infarcts, brain abscesses, multiple sclerosis (acute plaques), inter alia, also show contrast enhancement with gadolinium [13]. Furthermore, while the pattern of contrast enhancement is often useful in identifying tumors, this is not always the case. For example, ringlike enhancement, which is considered characteristic of brain abscesses, can also be seen with cystic brain tumors [24]. Higher-grade tumors tend to show more contrast enhancement, but this feature is not reliable for judging tumor grade. Low-grade pilocytic astrocytomas typically also enhance. The finding of a duralbased, extra-axial mass with dense, uniform contrast enhancement is characteristic of meningiomas [3]. On the other hand, for example, gliomatosis cerebri does not usually show a focal enhancing mass [3]. It should also be pointed out that malignant lesions from areas that are T2 hyper-intense, but outside the contrast-enhancing region have been frequently reported [25]. Vigneron et al. state [26], the contrast enhancing volume is normally considered to define the region of tumor{however} in practice, the contrast-enhancing region is often not an accurate reflection of the true extent of active tumor. This can be due to the presence of non-enhancing tumor and/or contrastenhancing necrosis. Although the lesion on the T2 weighted image is large and may encompass the tumor, it also reflects non-specific effects such as edema and inflammation. These factors produce considerable uncertainty concerning the reliability of MRI techniques for defining the treatment target and evaluating therapeutic success (p. 89). 8.2.3 Diffusion Weighted Imaging

Diffusion-weighted imaging (DWI) provides image contrast that depends upon the molecular motion of water. This technique has been useful in detecting acute ischemic stroke, since the loss of oxygen leads to increased signal amplitude in the affected tissue [27, 28]. DWI has been helpful in other disease of the brain, providing earlier detection of pathology, although the findings are not pathognomonic [29]. For example, Nakaiso et al. [30] found using DWI that both glioblastoma and brain abscesses appeared as hyper-intense lesions. DWI can help

84

differentiate tumor malignancy from secondary effects of treatment on the tumor [2, 31].

The current role of MRI and biopsy in brain tumor diagnostics is described succinctly by Howe and Opstad [32]: Accurate diagnosis is essential for optimum clinical management of patients with intracranial tumors. When accessible, most tumors are surgically resected, but there is a balance between removing as much tumor tissue as possible, whilst maintaining vital brain functions, and RT is often used to treat any remaining cancerous tissue. Currently there is widespread use of MRI for determining tumor extent for surgical and RT planning, as well as for post-therapy monitoring of tumor recurrence or progression to higher grade. MRI can provide an initial diagnosis of an intra-cranial mass lesion with a success rate of 30-90% depending on tumor type, but a biopsy is still generally considered the gold standard for determining the cancer type and degree of malignancy. However, a 1.7% mortality rate has been reported with biopsy procedures, and in a study of 550 patients undergoing stereotactic biopsy: 8% had abscesses or inflammatory processes, 2.2% had other lesions, 3.4% were nondiagnostic, and 8% suffered complications. A non-invasive and accurate prediction of lesion type would reduce unnecessary surgical biopsy procedures for non-cancerous lesions and for less accessible tumors that would be treated by radio- or chemotherapy rather than surgical resection (emphasis ours, p.123)[32].

8.3

Primary diagnosis of brain tumors using in vivo 1H-MRS-MRSI

The combination of anatomic plus molecular imaging via MRI together with MRS, yielding MRSI is particularly promising for brain tumor diagnostics, generating unique information [33]. As recently stated by Brando and Domingues [17], MRI has provided high temporal, spatial and contrast resolution methods to assess structure. However, the need to assess beyond the purely anatomic aspects, such as biochemistry and tissue physiology, required the development of functional techniquesAs a non-invasive method providing metabolic information about the brain, MRS enables tissue characterization at a biochemical level surpassing that of conventional magnetic resonance imaging. MRS is also able to detect abnormalities that are invisible to MRI, because metabolic abnormalities often precede structural changes (p. 2).

85

Each of the MR visible metabolites that provide some information for brain tumor diagnostics is reviewed Table 8.1. These are listed in ascending order of appearance on the chemical shift spectrum. For more detailed information on brain metabolites found on proton MRS, see [34] based upon in vitro data. 8.3.1. Comparisons with normal brain tissue As discussed, despite the fact that conventional MRI is considered the imaging technique of choice for evaluating brain tumors (p. 671) [14], many other non-malignant pathologies can mimic the MRI findings of intra-cerebral neoplasms. In this section, we will review the published findings on MRS and MRSI in the primary diagnosis of brain tumors, first presenting the diagnostic information provided by individual metabolites and then by metabolite ratios. In columns 4 and 5 of Table 8.1 the strength and consistency of the empirical data are summarized concerning the individual metabolites in the primary diagnosis of brain tumors. A summary comparison concerning metabolite ratios is presented in Table 8.2. Metabolite levels
Choline As discussed in Chapter 7, the hallmark for distinguishing neoplastic from non-neoplastic lesions using 1H MRS has been increased levels of choline; these are associated with high cell membrane turnover and high cell density, as occurs with proliferation of brain tumor cells. Sijens and Oudkerk [35]1 report a mean choline peak area of 34 3 in brain tumors (18 gliomas and 5 metastatic lesions) compared to 27 1 for unaffected brain tissue (p < 0.05). Significantly elevated choline in 12 patients with suspected brain tumors compared to levels in brain MRS recordings from 11 healthy volunteers was reported by Utriainen et al. [36]. However, choline can be low in very small brain tumors and with necrosis as in e.g. GBM. Brain tumors in children can also show low choline [17, 37]. Comparing MRS findings with histopathology, Dowling et al. [38] note that low choline could be found in mixed tissues that contained some brain tumor. A published case of gliomatosis cerebri showed normal choline levels [39].

Sijens et al. [35] used MRSI, 1.5T, PRESS, average of 5 voxels of normal tissue, TR=1500 ms, TE=135ms normalized to unaffected brain tissue, curve fitting with assumption of Gaussian line shapes, calculated metabolite area ratios.

Table 8.1

Individual metabolites from in vivo 1H MRS studies in brain tumor diagnostics: Summary of evidence
Brain Tumor versus Normal Brain Tissue Brain Tumor versus nonNeoplastic Lesions Brain Tumor Grading Brain Tumor Types

Metabolite: Assignment(s) [ppm] ++/Significantly higher lipid or presence with higher grade gliomas -+++

Neurobiological Significance/Age and Regional Variation

Methodological Issues for Neuro-oncology

Lipids (Lip) [0.9, 1.3 or 1.4]

+/Meningioma versus glioma

MRS appearance is usually pathological: liberation of triglycerides or long-chain fatty acids Significantly lipid or presence in tumor compared to healthy persons. Some overlap of ranges in 1 study Non-specific finding: In brain tumors but nearly as often with the same level in: Infarction / hypoxia Abscess Toxoplasmosis Tuberculoma Inflammation Multiple sclerosis

Suggests necrosis and/or disruption of myelin sheath

-Observed about as often in metastases as in higher-grade gliomas

20% of the dry weight of brain is lipid

Possible contamination from skull Overlap with lactate and alanine Seen mainly with short TE, although long T2 lipids also observed Overlap with amino acids at 0.9 ppm Difficult to distinguish from normally elevated baseline in this region

Lactate (Lac) [1.3] doublet

End product of failed oxidative metabolism / sign of anaerobic glycolysis

++/Significantly lactate or presence in tumor compared to healthy persons. Present in contra-lateral normal brain tissue in 1 study

-Non-specific finding: Seen in brain tumors, but also in

++ /Mainly corroborates histopathology

+/Gliomas versus Meningiomas

Small peak may be detected in newborns

Inverts below baseline at intermediate TE ( 140 ms) Overlap with lipids at short TE Requires proper shimming to appear as doublet Normal constituent of cerebrospinal fluid can appear spuriously with errors of voxel placement/localization

+/Metastases versus 1 brain tumors

Hypoxia, /ischemia Abscess Necrosis Epilepsy Demyelinating disorders

Metabolite: Assignment(s) [ppm]

Neurobiological Significance/Age and Regional Variation

Methodological Issues for Neuro-oncology

Brain Tumor versus Normal Brain Tissue

Brain Tumor versus nonNeoplastic Lesions

Brain Tumor Grading

Brain Tumor Types

Alanine (Ala) [1.48] doublet

Overlaps with lipids,

-Non-specific finding: Seen in brain tumors (especially meningiomas) but also in Abscess Neurocysticercosis

Levels are probably related to alterations in energy metabolism, with partial oxidation of glutamine, rather than glycolysis

Alanine levels may be overestimated if low lipid is present

+++/Meningiomas Mainly significantly or alanine more often present than in other brain tumors Mean [alanine] in meningioma < metastases in 1 study +/Metastatic versus 1 brain tumors +/ with higher grade tumors, except gliomatosis cerebri

Overlap with Glx at short TE +++/-

-Non-specific finding: in nearly any brain insult including malignancies in schizophrenia

N-Acetyl Aspartate (NAA) [2.02, 2.5, 2.6]

Can be spuriously with contamination from surrounding brain tissue

+ in meningiomas compared to tumors from glial precursors

Significantly in tumors compared to unaffected parts of the brain May be normal if tumor is small, lowgrade, or infiltrative

+/Metastatic versus 1 brain tumors

Marker of neurons & axons with neuronal injury or death (may be normal acutely) Can reappear with neuronal regeneration Can respond to metabolic events (e.g. hyperosmolarity) in hippocampus and cerebellum Progressive after birth, in the elderly

Metabolite: Assignment(s) [ppm] Brain Tumor versus Normal Brain Tissue Brain Tumor versus nonNeoplastic Lesions Brain Tumor Grading

Neurobiological Significance/Age and Regional Variation

Methodological Issues for Neuro-oncology

Brain Tumor Types

++
++

Glutamine/ Glutamate (Glx) [2.02, 2.5, 2.63, & 3.8]

--

Meningiomas: or often present compared to other brain tumors

+ 2.05 - 2.5

May be associated with accelerated cell proliferation Astrocyte marker, with hypoxia (reflects protective function of astrocytes) Participates in the redox cycle related to lactate accumulation (excitotoxin), converted via glutamine synthetase to glutamine

Significantly more frequent in high grade gliomas and metastases than in normal parts of the brain

++/-

3.6 - 3.8

In metastatic tumor significantly > Glx/Cr, in peri-tumoral region significantly < Glx/Cr compared to highgrade glioma,

Peaks are low despite high concentrations due to J coupling Requires short TE (3035 ms) to be identified Glutamine in and region, glutamine + glutamate in region If tumor treated with mannitol [peak = 3.8 ppm], can have confusion Can be over-estimated if glutathione at 2.9 ppm not included Overlaps with many other metabolites Non-specific finding: Seen in brain tumors, but also in Infarction Infection Bipolar disorders Attention-deficit hyperactivity disorder Hepatic encephalopathy Reyes syndrome Ornithine transcarbamylase deficiency

Glx more often present in meningiomas than metastases

+ described as an inconstant finding in brain tumors

Creatine (Cr)

-+/Non-specific finding: Can be in brain tumors, but also in infection, necrosis, stroke, acute multiple sclerosis with higher grade, except gliomatosis cerebri

Creatine + Phosphocreatine [3.02, 3.94]

Synthesized in liver & kidney (may in hepatic or renal disease) Key marker of cerebral E metabolism Internal referenceconsidered stable ( severe homeostatic compromise) in thalamus and cerebellum, grey matter in the elderly

Presence in meningiomas could be contamination from surrounding brain tissue

Meningiomas tend to show or absent Cr as nonneuronal tumors, compared to tumors from glial precursors

Brain Tumor versus nonNeoplastic Lesions

Brain Tumor Types

Metabolite: Assignment(s) [ppm] Choline (Cho) [3.22]

Methodological Issues for Neuro-oncology Entire signal contributes to the single narrow peak (No J coupling)

--

in meningiomas
versus other brain tumors +/Metastases vs 1 brain tumors (peri-tumoral choline consistent with 1 brain tumor)

Pathological changes amplified almost 10-fold May be low in very small tumors, in necrotic tumors, in pediatric intra-cerebral tumors, and in gliomatosis cerebri Report of significantly choline in highgrade compared to low-grade gliomas

Neurobiological Significance/Age and Regional Variation Marker for membrane damage, cellular proliferation and density From phospholipid metabolism of cell membranes, Sum of phosphorylcholine + glycerophosphorylcholine + 5% free choline in thalamus and cerebellum, in white compared to grey matter in infants and in the elderly Brain Tumor versus Normal Brain Tissue +++/Significant in brain tumors compared to unaffected brain tissue, & to healthy persons + Non-specific finding: Seen in brain tumors, but also in Infiltrative processes Ischemia Encephalitis Demyelinization Organizing hematoma Depression Schizophrenia Alzheimers disease Downs syndrome Diabetes mellitus Epilepsy Brain Tumor Grading ++/-High-grade gliomas tend to have choline, unless necrotic +/-

Myoinositol (mI) [3.56, 4.06]

Myelin degradation product Cell-volume regulator (osmolyte) in astrocytes

Requires short TE (3035 ms.) to be identified

--

Glycine peak also at 3.56 ppm Also reported to be

Reports of mI in various brain tumors,

Close to Glx peak

Non-specific finding: Seen in brain tumors, but also in Acute & chronic plaques of multiple sclerosis Alzheimers disease Bipolar disorders, Creutzfeldt-Jakob Leukodystrophies Downs Syndrome

+/Reports both of tendency for and mI with increased tumor grade (sometimes a mI-glycine peak is assessed)

++ Meningiomas: mI absent or lower compared to several other brain tumors +/Reports both of and mI in metastases compared to 1 brain tumors

Table 8.2

Metabolite Ratios from in vivo 1H MRS studies in brain tumor diagnostics: Summary of evidence
Distinction between Tumor and nonNeoplastic Lesions Tumor Grading +/++ Significantly in peri-tumoral region of metastases compared to high-grade gliomas +/Non-specific finding; can also be in Inflammation Stroke Multiple sclerosis Radiation necrosis Gliosis +/Histopathological Typing

Metabolite Ratios

Distinction between Tumor and Normal Brain Tissue

Cho/Cr

+++/ Several reports of in various tumors compared to normal, some statistically significant 1 report of similar mean levels and substantial overlap with healthy volunteers

Cho/NAA

+++

Several reports of in various tumors compared to normal, some statistically significant

+/Non-specific finding; can also be in Inflammation Stroke Multiple sclerosis Radiation necrosis Gliosis

Caption for Tables 8.1 and 8.2

+++ ++ + --

Evidence from several independent studies with statistical confirmation Evidence from at least 2 independent studies, with at least 1 statistical confirmation Observational and /or from published case reports only (at least 2) or from only 1 study of a series of patients Contrary observational and /or from published case reports (including overlapping ranges) Evidence to the contrary with statistical confirmation

91

N-Acetyl Aspartate Since NAA is a marker of number and viability of neurons, a normal level of NAA is generally considered to be incompatible with tumor [17]. Sijens and Oudkerk [35] report a mean peak area of 46 1 for NAA in unaffected brain tissue versus 13 2 at sites of brain tumors (significant difference at p < 0.01). However, with small tumors, or those that are low grade, the NAA level may be normal. NAA can sometimes also be normal with infiltrative brain tumors [17]. Assessment of the T2 relaxations times of NAA can be informative. These were found to be significantly lower in gliomas compared to normal volunteers in a paper by Isobe et al. [40]2. These authors also reported significantly lower NAA concentrations in the patients with gliomas. They note that shortening of T2 relaxation time is associated with limitation in molecular movement, or, conversely, that a long T2 indicates that the molecules move more freely. Their data suggest that the molecular motion of NAA and Cr is limited in these gliomas. Glutamine and Glutamate In a study by Fan et al. [41]3 Glx at 2.3-2.5 ppm was found significantly more often at the site of brain tumors (high-grade gliomas and metastases) among 22 patients compared to normal parts of the brain. These authors observe that glutamate may act as an excito-toxin associated with accelerated cell proliferation of malignant brain tumors. Howe and Opstad [32] note that Glx often appears in meningiomas and consider that this could be related to changes in energy metabolism with partial oxidation of glutamine rather than glycolysis, the end product being alanine, which is also elevated in meningiomas (see Section 8.5). Glutamine and glutamate are present in relatively large concentrations in all brain regions, but due to J coupling their peaks are relatively low. Glutamine is an astrocyte marker [17]. Lactate As discussed in Chapter 7, due to the predominance of anaerobic glycolysis, a lactate peak is seen often in tumors. Lactate is found in most pediatric brain tumors [17]. A lactate peak was detected in 3 of 10 patients with high-grade gliomas, as reported in a study by Saindane et al. [42]. While among the 18 patients with gliomas and 5 with metastatic brain tumors a mean level of lactate was registered, lactate was totally absent from all the unaffected brain tissue investigated by Sijens et al. [35]. In contrast, however, Tarnawski et al. [43]4 report that while the Lac/NAA ratio was significantly higher in the tumor bed of 51 adult patients with various grades of gliomas compared to the uninvolved lobe, lactate was
2

Isobe et al. [40] performed a single voxel study, 1.5T, PRIME (Proton Regional Imaging of Metabolites) sequence, variable TR and TE for calculation of T1 & T2 relaxation times, for calculation of concentrations TR/TE=2000/136 ms. 3 Fan et al. [41] applied MRSI, 2T, PRESS TR/TE=1500/45 ms, also TE=135ms for detection of inverted lactate peak, peak areas calculated. 4 Tarnawski et al. [43] used single-voxel MRS, 2T, PRESS, TR/TE=1500/35ms, automatic fitting software.

92

still present in some of the contra-lateral normal tissue. Among 30 healthy volunteers, however, no lactate whatsoever was detected. Lipids Depending upon the presence of necrosis, lipids may or may not be detected in brain tumors. Lipids are generally described as absent in normal brain tissue [17, 32]. In a study by Tarnawski et al. [43] there was no detected lipid whatsoever in brain MRS recordings from 30 healthy volunteers, and significantly higher lipid/NAA was found in the tumor bed of 51 patients (3.33 3.79) compared to 0.73 0.85 on the contra lateral, non-involved lobe. As shown in Table 8.3, based on the study of Smith et al. [44]5, while the lipid to creatine ratio was significantly lower among 5 healthy volunteer compared to 20 patients with brain-stem neoplasms, 2 of the 5 referents had some lipid detected as opposed to 18 of the 20 patients with tumors.

Table 8.3 Lipid presence and ratios to creatine in patients with brain stem tumors and healthy volunteers (From data of Smith et al. [44])
Lipid Peak detected at 1.3 ppm Patients with brain stem tumors Healthy volunteers 18 of 20 patients
Lipid/Cr Ratio
Mean Se6 (Range)

1.9 0.7 (0 8.0)

2 of 5 healthy persons

0.04 0.02 (0 0.12)

Myoinositol Since myoinositol is a precursor in lipid metabolism, it may be elevated with cell proliferation [32]. A number of reports indicate that myoinositol is elevated in various brain tumors [36, 39, 43, 45, 46]. However, Brando and Domingues [17] state just the opposite, namely, that intra-cerebral neoplasms are one of the conditions associated with decreased myoinositol.

Metabolite ratios comparing brain tumors and normal brain

In Column 2 of Table 8.2 the empirical data are summarized concerning the distinction between brain tumor and normal brain tissue using metabolite ratios.
5

Smith et al. [44] used both single voxel MRS and MRSI, magnetic field strength not specified, PRESS, TE=135 or 270ms (mainly 135 ms for referents, 270 ms for patients with brain tumors), TR=1500-1600ms, peak height ratios by measuring the highest point on each peak and normalizing to Cr. Lipids and lactate determined at the interval between 1.0 and 1.6 ppm. 6 Se denotes standard error of the mean.

93

In a comparison of MRI-localized 1H MR spectra from the tumor bed in 51 patients with high-grade gliomas with those from 30 healthy volunteers, significantly higher choline to creatine and choline to NAA ratios in the former was found by Tarnowski et al. [43]. These authors also found that the lactate to NAA, myoinositol to creatine and lipid to NAA ratios were significantly higher in the tumor bed, while the NAA/Cr ratio was significantly lower. Lin et al. [47] reported that the choline to creatine ratio was significantly higher at short TE among 49 patients with biopsy-proven brain tumors compared with 14 controls. The mean choline to NAA ratios from brain regions without tumor were 1.7 1.2 versus 5.3 2.1 to 6.8 3.7 in gliomas, depending upon the grade (all significant at p < 0.05), in the study by McKnight et al. [25], in which all findings were confirmed histopathologically. Increased choline and decreased NAA were found by Utriainen et al. [36] to distinguish brain tumors in 12 patients from corresponding regions in the normal brain recorded in 11 healthy referents. The Cho/NAA ratio is considered to be helpful in identifying low-grade neoplasms in children, in whom the choline concentration might be low or normal within the tumor. Since active myelination is on going in the normal parts of the brain, the choline to NAA ratio would be increased at the tumor site [17]. In contrast with these positive findings distinguishing brain tumors from normal tissue using ratios of choline to either NAA or creatine, in the study by Smith et al. [44] the range of choline to creatine ratios among 5 healthy volunteers was 1.54 2.14 (mean standard deviation (Sd) = 1.8 0.1), whereas among patients with solitary neoplastic brain stem lesions (astrocytoma, ganglioglioma, metastatic disease) the choline to creatine ratio was between 0.36 and 4.38 (mean Sd = 2.0 0.2).
Cut-off points for metabolite ratios In some studies metabolite ratios are treated as a dichotomous variable. McKnight et al. [25] used a cut-off value of 2.5 for the choline to NAA ratio, reporting that biopsy samples containing tumor (Grade II to Grade IV gliomas) were distinguished from those with normal, edema, gliotic and necrotic tissue with 90% sensitivity and 86% specificity, comparing pre-operative 3D MRSI and histopathology. These authors also reported that up to half of the T2 hyper-intense regions outside the contrast-enhancing lesion contained Cho/NAA > 2.5. Cho/Cr ratios 1.7 were reported by Murphy et al. [48]7 to be associated with unequivocal tumor presence, and that ratios 1.3 were seen in normal brain tissue. Among 31 patients with malignant brain tumors, all regions with confirmed cancer showed significant choline levels and a Cho/NAA >1.3 in an investigation by Vigneron et al. [26]8. As noted, however, in the study of Smith et al. [44], choline to creatine ratios in the brain of healthy persons were found to be as high as 2.14, while choline to creatine ratios as low as 0.36 could be seen at sites of brain tumors.

7 8

Murphy et al. [48] applied MRSI, 1.5T, PRESS TR/TE=2000/136 ms. Vigneron et al. [26] used MRSI, 1.5T, PRESS, TR=1000ms, TE=144 or 272ms. No correction for T2 metabolite relaxation times, spectral intensities of metabolites were quantified automatically.

94

8.3.2 Comparisons with non-neoplastic brain lesions The reader is directed to column five of Table 8.1 for a summary of the empirical data on how well individual metabolites detected using 1HMRS distinguish brain tumors with non-malignant brain lesions. Column 3 of Table 8.2 provides for a summary for metabolic ratios. Metabolite levels
Choline Choline is found to be elevated not only in brain tumors, but can also be increased in many other conditions [17]. These include: Sub-acute and chronic ischemia, Infiltrative processes, Encephalitis, Demyelinization (acute and in chronic plaques of multiple sclerosis), Organizing hematoma, Alzheimers disease, Downs syndrome, Diabetes mellitus, Depression, Epilepsy (reflects reactive astrocytosis). N-Acetyl Aspartate As a marker of neuronal viability and density, NAA can be decreased in almost any brain insult, although with acute damage NAA levels may still be normal. With maximally severe acute cerebrovascular accidents, a complete loss of NAA can occur [32]. NAA is found to be elevated in developing children, as well as with axonal recovery and the very rare Canavans disease [17]. With regeneration of neurons or axons, or the re-organization of brain tissue, NAA returns to normal. Metabolic events such as hyper-osmolarity can affect NAA concentrations [49]. Creatine Creatine levels reflect cerebral energy metabolism. These are considered to be relatively stable, with alteration usually reflecting severe compromise of homeostasis. For this reason, creatine is often used as an internal reference [49]. Decreased levels of creatine are very non-specific, and may occur with tumors, as well as infections, necrosis, stroke, acute multiple sclerosis, inter alia [17]. Creatine levels in brain tumors have been found to vary [49]. Glutamine and Glutamate Glutamine and glutamate levels are affected by diverse pathological processes [49]. As noted earlier, Glx was found in high-grade gliomas and metastases in the study of Fan et al. [41]. However, Brando and Domingues [17] consider that except for meningiomas, elevated Glx suggests non-neoplastic lesions. These include infection, infarction, hepatic encephalopathy, ornithine transcarbamylase deficiency, and bipolar affective disorder and attention-

95

deficit hyperactivity disorder. In Reyes syndrome elevated Glx is the main spectroscopic abnormality seen using 1H MRS [17].
Lactate As mentioned, lactate may be detected in brain malignancies, and is nearly always seen in pediatric CNS tumors. However, since it is an indicator of anaerobic glycolysis, the appearance of a lactate doublet on brain 1H MRS is not specific to tumors, and is commonly seen with the following conditions [17, 30, 42, 50]: Hypoxia, including ischemic stroke, Abscesses, Hydrocephalus, Necrosis, Cystic lesions, Epileptic seizure (lasting about 1 week thereafter), Tumefactive demyelinating lesions. A lactate peak was more common in tumefactive demyelinating lesions (4 of 6) compared to gliomas (3 of 10) in the study by Saindane et al. [42].

Lipids Since the appearance of lipids indicates necrosis or disruption of the myelin sheath, it is a highly non-specific finding. Besides brain tumors, lipids are commonly detected in the following intra-cerebral processes [17, 24]: Abscesses, Brain hypoxia / infarction, Demyelinization, Toxoplasmosis, Cryptococcoma, Epileptic seizure (lasting about 1 week thereafter).

Table 8.4 Lipid presence and ratios to creatine in patients with brain stem tumors and patients with non-neoplastic brain lesions9
(From data of Smith et al. [44]). Lipid Peak detected at 1.3 ppm Patients with brain stem tumors Patients with nonneoplastic lesions 18 of 20 patients
Lipid/Cr Ratio
Mean Se (Range)

1.9 0.7 (0 8.0)

13 of 14 patients

1.9 0.7 (0-4.0)

9 In the study of Smith et al. [44], 9 of the 14 recordings in patients with non-neoplastic lesions were with TE=270 ms, the other 5 were at 135 ms. In 17 of the 20 recordings of patients with tumors, TE=270 ms, and in the other 3 at 135 ms; see footnote 5 for further encoding details.

96

As seen in Table 8.4, based on the study of Smith et al. [44], the mean Lip/Cr was identical among 14 patients with various non-neoplastic brain stem lesions (fungus, gliosis, inflammation, hamartoma, multiple sclerosis and necrosis) compared to the 20 patients with brain-stem tumors. All but 1 of the patients with non-neoplastic lesions had lipid present [44].
Myoinositol As noted previously, some dispute exists concerning whether myoinositol is elevated in brain tumors, although several studies are affirmative. Myoinositol can also be increased in the following non-neoplastic brain disorders [17]:

Alzheimers disease, Multiple sclerosis (acute and chronic plaques), Epilepsy (indicating reactive gliosis and astrocytosis), Encephalitis, Bipolar affective disorders, Creutzfeldt-Jakob disease, Leukodystrophies. The fact that myoinositol acts as both an osmolyte regulator and as an astrocyte marker helps to elucidate these seemingly diverse findings concerning this metabolite [49]. Alanine Alterations in energy metabolism, with partial oxidation of glutamine, rather than glycolysis, are considered the likely reason for elevated alanine levels, as well as Glx, seen in meningiomas [32]. Non-neoplastic brain lesions associated with increased alanine include [17]: Brain abscesses, Neurocysticercosis.
Acetate and succinate Brain abscesses often show acetate [1.92 ppm] and succinate [2.4 ppm], whereas this is generally not the case for brain neoplasms [17, 24, 51]. The appearance of these metabolites is thus considered to distinguish brain abscesses from cystic or necrotic tumors [17].

Metabolite ratios comparing brain tumours & non-neoplastic lesions In some studies, metabolite ratios were found to distinguish brain tumors from non-neoplastic lesions. High choline, low NAA together with a minimal lactate peak on proton MRS were helpful in differentiating bilateral thalamic astrocytomas from encephalitis and neurometabolic disorders in a study by Gudowius et al. [22]. As noted, these disorders are difficult to distinguish among children on the basis of MRI. Similar findings were reported by Moller-Hartmann et al. [52]; high choline with low creatine and NAA was characteristic of

97

brain tumors, while non-neoplastic lesions such as brain abscesses or cerebral infarction showed decreased choline, creatine and NAA.
Null results Metabolite ratios did not consistently identify which lesions were brain tumors and which were not. In the study of Smith et al. [44] there was substantial overlap in choline to creatine ratios among normal, non-neoplastic lesions and tumors. Cho/Cr ratios in non-neoplastic brain stem lesions (fungal infection, gliosis, inflammation, multiple sclerosis and necrosis) ranged from 0 - 2.6, (mean Sd = 1.4 0.2), compared to the range in malignant brain stem tumors of 0.36 to 4.38 (mean Sd = 2.0 0.2). Multiple sclerosis and brain tumors may have similar features on MRI, as noted earlier. Saindane et al. [42] compared 6 patients with tumefactive demyelinating lesions (TDL) and 10 patients with high-grade gliomas. These lesions were not distinguishable on MRI. Moreover, there were no significant differences in mean Cho/Cr in contrast-enhancing, central or peri-lesional areas of TDL versus glioma. The mean central-region NAA/Cr in gliomas was significantly lower than in TDL, but mean NAA/Cr in other regions was not significantly different. A lactate peak was seen in 4 of the 6 patients with TDL and in 3 of the 10 with glioma. These authors [42] conclude: overall metabolite profiles of both lesions were similar{There is a} need for the cautious interpretation of spectroscopic findings (p. 1378). Gliomas and acute multiple sclerosis plaques may have indistinguishable MR spectra. Butteriss et al. [53] describe a patient in whom elevated choline, and decreased NAA were found, together with increased lactate and lipid - a pattern that could be either low-grade glioma or acute demyelinating plaque. The repeated MRS revealed no change; this was considered to be incompatible with the natural history of an acute plaque and low-grade glioma was diagnosed. Surgical removal confirmed that this was an oligodendroglioma.

The problems of relying upon ratios of choline to creatine and to NAA are underscored by Howe and Opstad [32]. These authors conclude that although elevated Cho/Cr (and elevated choline if there is no significant necroses) and reduced NAA are found in brain tumors, these characteristics must be used cautiously, since other lesions may exhibit similar features.

Non-neoplastic brain lesions that may have tumorlike in vivo 1H-MR spectra:

Inflammation, Cerebrovascular accidents, Multiple sclerosis, Radiation necrosis, Gliosis.

98

Particularly difficult CNS pathology to distinguish from brain tumors with in vivo 1H-MRS:

Inflammatory pseudotumor, Organizing hematoma, Sub-acute encephalitis due to herpes virus type 6, Fulminant acute demyelination.

On the other hand, Murphy et al. [54] cite several individual cases in which in vivo 1H-MRS made a special contribution to identifying and characterizing brain tumors that were otherwise difficult to distinguish from non-neoplastic lesions including infection, stroke and arachnoid cyst.

8.4 In vivo MRS and MRSI for brain tumor grading

As explained in Section 8.1, the grading of brain tumors is of utmost importance with respect to treatment and prognosis. To illustrate, patients with low-grade astrocytoma who receive combined therapy have approximately a 65% 5-year survival, whereas for similar treatment strategies among patients with malignant astrocytoma, approximately 65% percent of those age 15 to 44 survive 17 months. For those age 65 with malignant astrocytoma, the mean survival is five months [32]. Accurate grading of intra-cerebral neoplasms can be very difficult, due especially to tumor heterogeneity. For this reason in particular, brain biopsy may not provide the definitive answer, besides being associated with substantial morbidity. Thus, there has been interest in using in vivo MRSI with full volumetric coverage for grading brain tumors. Results using 1H MRS and MRSI Individual Metabolites and their ratios
Choline, NAA, Creatine The greatest reduction in NAA and Cr compared to lower grade tumors is usually seen in higher-grade gliomas [17, 40]. Ratios of choline to NAA or creatine are often elevated in higher-grade tumors [55, 56]10. Glioblastoma multiforme has been characterized by lower NAA and creatine, and higher choline levels in comparison to low-grade astrocytomas [57]. Choline levels were significantly lower among 15 children with low-grade neuroglial tumors compared to 6 children with high-grade tumors studied by Tzika et al. [58].11

10 11

Li et al. [56] used 1.5T, MRSI, PRESS, TR/TE=1000/144 ms. Peak heights. Tzika et al. [58] used 1.5T, MRSI, PRESS, TR=333 ms, TE=40ms.

99

On the other hand, while higher mean choline levels and lower mean NAA levels are generally associated with high-grade tumors, there are large standard deviations, which Vigneron et al. [26] consider to preclude accurate tumor grading. This is due, at least in part, to the fact that high-grade brain neoplasms frequently contain necrosis, such that choline levels and their ratios to NAA and creatine may be similar to or even less than those of low-grade tumors [17, 40, 56, 59]. For example, in a study by Isobe et al. [40], choline concentrations were found to be significantly lower in 13 high-grade compared to 10 low-grade gliomas. Some particularly illuminating data are provided by Li et al. [56]. These authors found when assessing T2 hyper-intense lesions that the mean (but not the median) Cho/NAA and Cho/Cr peak height ratios were somewhat greater in Grade IV Gliomas compared to those of Grades II and III. However, contrast-enhancing lesions from Grade III gliomas showed a mean Cho/NAA of 11.46 11.4 (median 7.5) versus 3.09 3 (median 2.26) in Grade IV gliomas at sites of contrast-enhancement. Similar, but less marked trends were seen for the Cho/Cr ratios. In contrast to these findings from voxels chosen by morphologic criteria from MRI findings, when metabolic criteria were used, a very clear relation between Cho/NAA and Cho/Cr ratios and each tumor grade appeared. Namely, when the regions of maximum Cho/NAA or Cho/Cr were chosen, the mean and median of the ratios clearly increased according to tumor grade. Moreover, compared to the ratios at sites of T2 hyper-intensity the Cho/NAA and Cho/Cr ratios were several times higher at the maximum sites. These findings for Cho/NAA ratios are graphically displayed in Figure 8.1.

Figure 8.1: Glioma tumor grade and Cho/NAA according to MRSI versus MRI based localization (From data of Li et al. [56])

14 12 Choline to NAA 10 8 6 4 2 0 Max Cho /NAA CE T2 Hyperintense Grade IV Grade III Grade II

Max Cho/NAA denotes area of maximum choline to NAA ratio detected using MRSI, CE denotes areas of contrast enhancement on MRI (Grade II Gliomas did not show contrast enhancement), T2 Hyperintense denotes T2 hyper-intensity on MRI

100

The study of Li et al. [56] clearly illustrates the importance of volumetric coverage with MRSI for CNS tumor diagnostics, especially for tumor grading. As stated by Spielman [60], metabolically abnormal regions may not precisely correspond to much larger T2 abnormalities. Thus, he concluded: sampling the most metabolically abnormal tissue may be a better choice than just sampling the tissue with the highest level of contrastenhancement (p. 31). The weakness of single voxel techniques that provide one spectrum which is presumed to be representative of the entire tumor is also pointed out by Vigneron et al. [26]. They note that this technique provides no information on the extent of the metabolic abnormality and suffers both from the inaccuracy of conventional MRI to define the exact extent of solid neoplasms and the considerable tissue heterogeneity within the tumor mass (pp. 98-99). Moreover, these authors [26] observe that high NAA can be due to contamination from adjacent tissues, especially in studies using a large single voxel. Thus, there is a definite need for 3D spectroscopic imaging for identification of viable tumor on the basis of both morphologic and metabolic parameters for targeting and following local therapy (p. 100). Lactate Lactate levels are generally found to parallel tumor grade [17, 51, 61-63]. This is biologically plausible, since the appearance of lactate caused by anaerobic glycolysis is the main metabolic pathway for glucose utilization by anaplastic brain tumors [64]. Histopathologic evidence is also corroborative, in that a positive correlation was found between the lactate peak and the degree of heterogeneity of the nuclear roundness factor in anaplastic gliomas [65]. However, in a study of 36 patients with brain tumors, Kugel et al. [66] found no correlation between tumor grades and lactate level. Lipids We have earlier noted that necrosis is characteristic of higher-grade brain tumors. This is manifested spectroscopically by the presence of lipid. As stated by Murphy et al. [54]: the spectra of glioblastomas are dominated by huge lipid - lactate resonances, and with few other features (p. 330). Several investigations show that the presence or greater concentration of lipids is associated with higher-grade intra-cerebral tumors [52, 56, 59, 67]. Myoinositol A trend towards lower myoinositol levels in higher-grade gliomas compared to those with a lower grade has been reported in several papers [46, 55, 68, 69]. In the study by Majs et al. [70a]12 however, while not statistically significant, the mean normalized area values of the glycine-myoinositol component were the highest for 25 patients with glioblastoma compared to those with anaplastic astrocytomas and low-grade astrocytoma. Moreover, mean myoinositol to creatine ratios were significantly higher in 14 patients

12 Majs et al. [70a] used 1.5T, PRESS, single voxel, TR/TE=2000/136 ms, MRUI [70b] and VARPRO (variable projection method) non-linear least squared fitting; fitted peak areas normalized to square root of the sum of squares of NAA, Cr and Cho.

101

with high-grade gliomas compared to 8 patients with metastatic lesions reported by Fan et al. [41]. Glioblastoma cerebri This unusual glial sub-type with very poor prognosis is seen to have nonsignificantly higher myoinositol than low-grade gliomas. This finding is considered to be related to glial activation [71]. This tumor infiltrates the brain lobes with atypical glial elements. It may infiltrate large areas of brain parenchyma, usually does not show contrast enhancement and is often indistinguishable from low-grade glioma on stereotactic biopsy. It is critically important to identify gliomatosis cerebri because of its poor prognosis, and poor response to RT and chemotherapy. Besides a non-significant elevation in myoinositol, gliomatosis cerebri has been distinguished from low-grade gliomas by elevated creatine and NAA, and often lower choline [71].

Combined Analysis A combination of metabolites was used by Herminghaus et al. [64]13 to perform tumor grading. With single-voxel 1H-MRS, linear discriminant analysis based upon normalized NAA, total creatine, choline, lactate and lipid to total Cr ratio was used to classify tumor grades among 90 patients with histopathologically confirmed glial brain tumors. Correct tumor grading was achieved in 77 - 94% of cases. The most serious grading errors occurred for altogether 2 patients with Grade I/II who were considered to be Grade III/IV and for altogether 3 patients with Grade III/IV who were assessed as I/II. These authors emphasize the importance of early detection of transformation to a higher tumor grade. They consider overall that spectral pattern analysis provided a high level of accuracy with the possibility of identifying cases that cannot be reliably graded, and thereby avoids therapeutic decisions based on ambiguous data (p. 79).
31

P MRS

31

P MRS was used by Maintz et al. [72] to compare patients with Grade II gliomas and glioblastomas to normal cerebrum. Table 8.5 shows some mean data that might be helpful for tumor grading14. Namely, the mean pH appears to increase from the normal group, to Grade II glioma to Glioblastoma. In addition, the mean PME to ATP ratio is greater for glioblastomas than for the Grade II Gliomas and normal group, for whom this ratio is very similar.

Herminghaus et al. [64] used 1.5T, single-voxel, PRESS, TR/TE=1500/135 ms, normalization as a ratio to total creatine (creatine plus phosphocreatine) obtained from the reference spectrum. 14 Statistical analysis was performed using the non-parametric Mann Whitney test, comparing each tumor group to the volunteers. Since the data for individual patients are not given and we cannot assume a normal distribution of the metabolite ratios, it is not possible for us to statistically assess differences between the patient groups.

13

102

Table 8.5 31 P MR spectral differences by glioma grade

(From data of Maintz et al. [72]) Glioblastoma (16 patients)
Mean Sd

Glioma Grade II (15 patients)

Mean Sd

Normal Brain (36 volunteers)

Mean Sd

PME/ATP

0.47 0.06 7.12 0.02

0.40 0.07 7.09 0.02

0.41 0.04 7.04 0.01

pH

The findings in the literature concerning pH in relation to brain tumors and therapy are noted to be very contradictory [72].

8.5 MRS & MRSI for histopathologic classification

MRS and MRSI have been used to help classify brain tumors. Del Sole et al. [73] consider MRS to be particularly appropriate, noting: whereas anatomical imaging is aimed at the diagnosis of brain tumors, biochemical imaging is better suited to tissue characterisation (p. 1851). We will focus here upon two of the most clinically important distinctions: between meningioma and other brain tumors especially those from glial precursors, and between metastatic disease and 1 brain tumors. Meningioma versus other brain tumors Meningiomas have a number of characteristic morphologic features that are helpful for identification on MRI, as noted in Section 8.2. These include the so-called tail-sign (dural enhancement pattern with contrast infusion adjacent to the tumor base), well-demarcated mass configuration, and certain typical locations (para-sagittal interhemispheric, lateral convexity, inter alia). Calcifications occur in approximately 25% of meningiomas and are best identified with CT [13]. While typical, these MRI signs do not provide absolute diagnostic accuracy. Given the usual slow growing nature of meningiomas, with rare malignant transformation, it is very important to distinguish these from other, more aggressive brain tumors, especially those of glial origin. A number of spectral features of

103

meningiomas assessed using in vivo MRS and MRSI are helpful in making this distinction. Individual Metabolites and their ratios
Alanine When alanine is found on MRS or MRSI, this is considered by some authors to be highly suggestive of meningioma [17, 49]. Treated as a dichotomous variable (present or absent), alanine was reportedly present in 15 of 19 cases of meningiomas, but was not detected in 11 Schwannomas, or in any of the 8 metastatic tumors examined in Ref. [74]15. Higher alanine (as well as glutamine/glutamate, see next sub-section), best discriminated between 22 patients with meningiomas and 9 with primitive neuroectodermal tumors in a study by Majs et al. [75]. This group of authors in another paper [76]16 reported that alanine was elevated in atypical meningiomas, compared to other tumors types. In a third paper, Majs et al. [70a] described alanine as the most characteristic resonance of meningiomas. Alanine was rarely seen in non-meningeal tumors in these studies. In contrast, Howe et al. [77]17 report that mean alanine concentrations were lower in 8 patients with meningiomas (2.0 1.9) compared to those of 6 patients with metastatic brain lesions (3.0 3.8). Glutamine-Glutamate and Glutathione Besides alanine, the appearance of Glx is also described as highly suggestive of meningioma [17]. In Ref. [74], increased Glx was detected in 12 of the 19 patients with meningiomas, but this was also found in 4 of 8 patients with metastatic brain tumors, though in none of the patients with Schwannomas. Majs et al. [70a] reported increased Glx in the patients with meningiomas compared to astrocytomas. As mentioned, these authors found that, in addition to alanine, Glx was the best metabolite for distinguishing patients with meningiomas from those with primitive neuroectodermal tumors. Opstad et al. [78]18 report that 6 patients with meningiomas had elevated Glx concentrations compared to 6 patients with astrocytomas as well as in comparison to normal white matter. The reduced glutathione resonance at 2.9 ppm was also significantly greater in the patients with meningioma compared to those with astrocytomas, as well as in the normal white matter in that study [78] in which a short TE was used. These authors note that glutathione has a major role in removing
15 Cho et al. [74] used 1.5T, single voxel, STEAM, TR/TE=2000/34 ms, and PRESS, TR/TE=2000/288 ms, relative peak heights. 16 Majs et al. [76] applied 1.5T, single voxel, PRESS, TR/TE=2000/136 ms, MRUI [70b] and VARPRO-non-linear least squared fitting, fitted peak areas normalized to square root of the sum of squares of NAA, Cr and Cho. 17 Howe et al. [77] used 1.5T, single-voxel, STEAM (TR/TE = 2000/30 ms, and PRESS (TR/TE = 2000/136 ms), AMARES (Advanced Method for Accurate Robust and Efficient Spectral fitting) + MRUI, excluded the first data point of the metabolite FID to minimize incorporation of large broad baseline components to avoid overestimation of metabolite signals (p. 224), no T1 or T2 corrections. 18 Opstad et al. [78] applied 1.5T, single-voxel, STEAM, TR/TE=2020/30 ms, LCModel (Linear Combination of Model in vitro Spectra) fitting.

104

free radicals and toxins form normal tissues, but in tumor cells it hampers effectiveness of anti-cancer therapy. They suggest: the ability to noninvasively quantify reduced glutathione may aid selection of treatment and also provide an indication of tumor aggressiveness (p. 632). Creatine and NAA Meningiomas are non-neural tumors. For that reason, NAA and creatine are described as being virtually zero, such that the presence of these metabolites would indicate contamination from surrounding brain tissue [17]. Though present, mean creatine and NAA levels were marginally lower in 8 patients with meningiomas compared to those from patients Grade II astrocytomas, anaplastic astrocytomas and glioblastoma multiforme in the study by Howe et al. [77]. Similarly, Majs et al. [70a, 76] reported lower creatine and NAA in meningiomas compared to astrocytic tumors. Moller-Hartmann et al. [52] reported absent NAA in meningiomas and neurinomas. A substantial reduction in phosphocreatine to creatine ratio was also reported in meningiomas assessed using 31P MRS [72]. As discussed in the next subsection, very low NAA and creatine may also be found in metastatic brain lesions. Myoinositol None of the 19 patients with meningiomas showed increased myoinositol compared to 10/11 of the patients with Schwannomas examined in [74]. Mean levels for a combined myoinositolglycine peak (mIG) were lower among the 8 patients with meningiomas than in glial tumors, metastases as well as in normal white matter, as reported by Howe et al. [77]. Furthermore, meningiomas were distinguished from hemangiopericytomas, which have a similar MRI appearance, but require different therapeutic strategies, by the presence of a larger peak at 3.56 ppm in the latter, attributed to myoinositol [45]. Choline High choline levels were used to help distinguish meningiomas from other brain tumors in Ref. [70a]. Mean choline levels were also found to be highest among 9 patients with meningiomas compared to those with gliomas or metastases in the study by Howe et al. [77]. Lipids The spectra of meningioma may show only lipids, according to Brando and Domingues [17]. On the other hand, in several studies minimal or no lipids were detected in meningiomas [70a, 74, 76].

Combined Analyses Meningiomas were most clearly distinguished from Schwannomas and metastases by the presence of alanine, lack of lipids and of myoinositol, and by the presences of a signal at 3.8 ppm at long TE in Ref. [74], in

105

which spectral characteristics were assessed dichotomously as present or absent. The most salient metabolic characteristics of the 29 meningiomas studied by Majs et al. [70a] were: alanine, glutamine-glutamate, compared to low-grade astrocytomas and anaplastic astrocytomas, choline compared to low-grade astrocytomas, glioblastomas and metastases, Relative in creatine compared to astrocytic tumors, Low or absent lipid at 1.3 ppm, compared to glioblastoma and metastases.

Majs et al. [70a] used a stepwise algorithm based upon these findings. The accuracy with which these authors identified meningiomas with single-voxel proton MRS, ranged from 83% (versus anaplastic astrocytoma) to 95% (versus low-grade astrocytoma). A total of 4 of the 81 cases meningiomas could not be classified to distinguish them from glioblastoma multiforme or metastases, and in 12 of 166 cases there was a misclassification of meningiomas versus the other tumors using this algorithm. In another investigation coming from three different centers19 and including 144 patients with histopathologically diagnosed brain tumors, the diagnostic accuracy of an automatic classification system at short TE was assessed [79]20. In this study there were 37 patients with meningiomas; these were identified by elevated intensities from regions around 3.7 ppm (glutamate, glutamine, alanine) and 2.33 ppm (glutamine, glutamate, macromolecules). Two of the 37 patients with meningioma were misclassified as having aggressive tumors (glioblastoma or metastases), and 2 of the 89 patients with aggressive tumors were misclassified as having meningiomas. There was no diagnostic confusion whatsoever between the meningiomas and 18 Grade II astrocytomas using the mentioned stepwise algorithm.

19 20

Including Barcelona, the center of Majs et al. [70b]. Tate et al. [79] used 1.5T, single voxel, STEAM or PRESS, TR=1600-2000ms, TE=20-32ms.

106

Distinction between primary brain tumors and metastatic lesions Brain metastases and primary brain tumors have a number of important clinical distinctions, particularly with respect to therapy and prognosis (see Section 8.1). One of the major distinguishing features of metastatic lesions is that they are often multiple, in about 70% of patients [13]. Their location is also quite characteristic. As noted, the anatomic distribution of metastatic brain tumors generally parallels regional cerebral blood flow, with common locations being the grey matter-white matter interface, and the border between the middle and posterior cerebral artery distributions. Unlike primary gliomas, brain metastases usually, but not always, appear on MRI as wellcircumscribed, nodular masses, clearly distinguished from the surrounding brain tissue. Metastases may contain central areas of necrosis. Minimal glial reaction is generated by metastatic lesions. Instead, a surrounding area of vasogenic edema is often described [13]. However, the relative cerebral blood volume in the peri-tumor region was found to be higher for high-grade gliomas than for metastatic lesions, in a study by Law et al. [80] using perfusion-weighted MR data. Some in vivo spectroscopic findings of metastatic brain lesions help to distinguish them from primary brain tumors. Individual Metabolites and their ratios
Choline, NAA and Creatine Peritumoral region Choline is usually low in the area surrounding metastases, since the peritumoral region of metastatic lesions is typically comprised of vasogenic edema rather than cellular infiltration. Elevated choline in the peri-tumoral region suggests a primary brain neoplasm [17]. Significantly lower choline-tocreatine ratios in the peri-tumoral region in 8 patients with metastases compared to 14 patients with high-grade gliomas were found by Fan et al. [41]. These authors term infiltration of adjacent brain parenchyma a unique feature of high-grade glioma. Law et al. [80] 21 report similar findings. Tumoral region Choline-to-creatine ratios were also significantly higher in the actual gliomas in these two studies [41, 80]. However, Fan et al. [41] note that high choline was also found in 4 of the patients with metastases suggesting that a high

21

Law et al. [80] used 1.5T, MRSI, PRESS, TR/TE=1500/135ms, maximum values identified from spectral maps, peak area ratios.

107

choline concentration in tumoral regions is characteristic of rapidly growing tumors, rather than unique for glioma (p. 81). Brando and Domingues [17] state that absent or nearly absent NAA and creatine levels are typical of metastatic brain lesions. However, the mean NAA levels from 6 patients with metastases were actually higher than those from 5 patients with low-grade astrocytomas, and 13 patients with glioblastoma multiforme in a study by Howe et al. [77]. Lipids Lipids are frequently detected in metastatic lesions, although, as discussed in the Section 8.4, these are also seen in high-grade gliomas [52, 70]. Long T2 lipids were observed in all 8 patients with metastatic lesions, and in 1 of the 19 patients with meningiomas, in Ref. [74]. Similarly, Fan et al. [41] reported lipids detected at around 1.3 ppm in all 8 patients with metastases. Lipids were also present for all 6 patients with glioblastomas and in 2 of 8 with anaplastic astrocytomas in Ref. [41]. Glutamine In 4 of the 8 patients with metastatic lesions glutamine was present, compared to 12 of the 19 patients with meningiomas, in Ref. [74]. Mean glutamine to creatine ratios were significantly higher in the tumoral and significantly lower in the peri-tumoral regions among patients with metastases compared to those of patients with high-grade gliomas in Ref. [41]. Alanine Findings concerning alanine are contradictory. Alanine was reported absent in all 8 patients with metastases examined in Ref. [74]. On the other hand, as noted earlier, mean alanine concentrations were reportedly higher among 6 patients with metastatic brain lesions compared to those with glial tumors as well as meningiomas in Ref. [77]. Myoinositol Of the 8 patients with metastatic brain lesions in Ref. [74], none showed increased myoinositol. Mean myoinositol to creatine ratios in the tumoral regions were significantly higher in 14 patients with high-grade gliomas compared to 8 patients with metastatic lesions reported by Fan et al. [41], as previously noted. In Ref. [70a], though not statistically significant, the mean normalized area values for the myoinositol-glycine resonance were the lowest in the 27 patients with metastases and the highest in the 25 patients with glioblastoma compared to all other patient groups with brain tumors. However, in Ref. [77], metastatic lesions showed a higher mean myoinositol plus glycine peak compared to patients with glioblastoma, although the levels in brain metastases were lower than those with lower grade astrocytomas.

108

Combined Analyses In Ref. [70a] patients with metastatic lesions and glioblastomas were grouped together. Comparisons were made between this group and patients with meningiomas, low-grade astrocytomas and anaplastic astrocytomas. Majs et al. [70a] also tried to distinguish glioblastomas from metastatic brain lesions, but found all the differences in normalized area values to be non-significant. Nevertheless, the authors reported that the myoinositol-glycine resonance was the most helpful in making this distinction.

8.6 MRSI for localization of brain tumors

Clearly, maximum precision in localization is vital for treating tumors affecting the brain. This is facilitated considerably by MRSI. MRSguidance based upon areas of increased choline to creatine ratios has improved the yield of diagnostic tissue obtained with stereotactic biopsy [81, 82]. It has been suggested that by using multivoxel MRS the area of highest choline can be identified, and that would be the ideal site for biopsy (p. 154) [17]. The feasibility of using MRSI for intra-operative decision-making was shown by Liu et al. [83]. Determination of target is vital for planning RT. Pirzkall et al. [84] suggest that the incorporation of MRSI into RT planning for high-grade gliomas could improve control while reducing complications (p. 915). Currently, the clinical target volume for RT of gliomas is generated by adding uniform margins of 2-3 cm to the area of T2 hyper-intensity [84, 85]. This is because the peri-tumoral uncertain zone, while appearing normal on MRI, can, in fact, be infiltrated by tumor when examined histopathologically [41]. With the help of MRSI to determine areas of high Cho/NAA, the shape and size of the clinical target volume could be optimized, with more confident sparing of un-involved brain tissue [84, 85]. Maps of Cho/Cr as well as Cho/NAA ratios have been helpful in refining RT dose contouring, as well as for surgical planning to treat brain tumors more effectively [32]. As noted, even lower grade gliomas are frequently infiltrative with poorly defined margins. The degree of tumor infiltration for each grade of glioma was found by Croteau et al. [37] to be best defined by the ratio of mean aggregate choline concentration at the tumor to creatine concentration in the contralateral, normal brain tissue. The current limitations of spatial resolution of MRSI, and the need for prospective studies to determine its actual clinical effectiveness in

109

determining boundaries of high- as well as low-grade gliomas, are stressed by Orszagh and Ostertag [86].

8.7 Gauging response to therapy by MRS & MRSI

The concentrations of a number of metabolites can suggest whether or not treatment of a brain tumor has been successful. Usually, the choline levels fall, whereas lipids and lactate may rise [32, 87, 88] with a good therapeutic response. In a study [89] of 24 children with brain tumors, those who responded to chemotherapy or RT also showed significantly lower choline and higher total creatine. Lipids and lactate, however, were lower among the responders than among those treated only by surgery or who did not respond to treatment. The only metabolite to independently predict active tumor growth was total creatine. The peri-tumoral region showed significant differences in Lip/Cr and Lac/Cr between patients with and without recurrent gliomas, in a study by Walecki et al. [90]. This region appears unchanged on standard MRI, as noted earlier. Favorable treatment response of patients with brain tumors has been found to show an increase in water diffusion. This coincides with cytotoxic cell eradication and preceded volume reduction [91]. Recognizing tumor recurrence as opposed to response to therapy Brain tumors frequently return after treatment, and often do so at a higher grade. In addition, the treatment itself, especially RT, provokes changes in brain tissue that are difficult to distinguish from tumor recurrence using MRI. Radiation necrosis, local response to immunotherapy, as well as the tumor itself all can show contrast enhancement. MRSI improves specificity by differentiating radiation necrosis or enhancement phenomena after local immuno-therapy, from tumor recurrence [32, 92-94]. Using the ratio of choline at the biopsy site to creatine in normal brain tissue > 1.3, Rabinov et al. [95]22 successfully identified glioma as opposed to radiation effects in 16 of 17 cases. In a study by Ng et al. [96] patients with stable disease after RT of cerebral gliomas showed lower choline to creatine ratios (mean 1.2, range: 0.4 2.1) compared to those with tumor recurrence (mean 3.0, range: 0.9 9.5). However,
22

Rabinov et al. [95] used 3T, PRESS, MRSI, TR/TE=1500/144 ms, peak areas determined by an iterative fit assuming Lorentzian line shapes.

110

there was considerable overlap between these two groups and therefore Cho/Cr ratios were considered to be inconclusive for determining if tumor re-growth or a stable process was occurring (p. 708). Further complicating the picture is that choline levels may also increase during RT, associated with gliosis [17]. More timely recognition of tumor recurrence It is known that MRS and MRSI provide greater sensitivity in detecting brain tumor recurrence compared to MRI alone. The choline levels, or choline to creatine ratios generally rise before the appearance of contrast enhancement [32, 87]. However, normal choline levels were reported in 6 of 15 patients with high-grade gliomas treated with RT, even though the tumors recurred within 3 months [97].

8.8 Prognostic information provided by MRSI

MRSI was reported to predict tumor-related length of survival in 60 patients with supratentorial gliomas in a paper by Kuznetsov et al. [98]23. The accuracy of this prediction was comparable to that based upon clinico-pathological features, including tumor grade. MRSI predicted tumor-related survival with more accuracy than did surgical debulking. In this study [98], the significant prognostic predictors from MRSI were that

Maximum Cho/Cr (p < 0.0001), Maximum Lac/Cr (p < 0.0001), Low NAA/Cr voxels24 (p < 0.002).

Tzika et al. [99] also suggest that in vivo MRSI can be used as a prognostic indicator. They examined 27 children with neuroglial brain tumors, and found that the percent change in choline to NAA ratio on MRSI, as well as relative tumor blood volume were significant predictors of tumor progression.

23

Kuznetsov et al. [98] used 1.5T, MRSI, PRESS, TR/TE=2000/272 ms (water suppressed), TR/TE= 850/272 ms (non-water suppressed), to correct for magnetic field inhomogeneities; the water suppressed data were divided by the unsuppressed ones after zero-filling the latter. Metabolite resonance intensities were determined as the area of Gaussian line shapes fitted to the resonance peaks relative to a baseline computed on the basis of a moving average of the noise regions of each spectrum. 24 The number of voxels with NAA/Cr < 2/3 of the normal mean.

111

8.9 2D J-resolved and in vitro MRS: Further insight into the metabolic characteristics of brain tumors
Localized 2D J-resolved MRS and in vitro MR spectroscopy provide further insights into many aspects of brain tumor diagnostics. Overlapping resonances in the lipid / lactate region In vivo 2D J-resolved MRS With localized 2D J-resolved MRS, Thomas et al. [100] were able to identify and quantify the lactate peak in the presence of overlapping lipid in a patient with glioblastoma. They conclude: in vivo J-resolved plots of human brain tumors indicate the exciting potential of this technique in extracting additional information from the conventional MR spectrum (p. 459). In vitro MRS studies The content of the resonances at 0.9 and 1.3 ppm is further clarified by in vitro MRS. The ratios between the 0.93 and 1.3 ppm peaks have been highly variable with in vivo MRS [101]. High-resolution studies of tumor cells show that the CH3 signal at 0.9 ppm arises from protein residues at 0.92 ppm and from lipids at 0.88 ppm. This is due to a nonlipid contribution super-imposed on the 0.9 ppm resonance. The large lipid content of the normal human brain is not visible with in vivo MRS. When lipid peaks are seen, this is clinically important, reflecting necrosis from any cause. Identification of specific lipids has heretofore required the high resolution of in vitro MRS applied to biopsy specimens from brain tumors [49]. In vitro MRS reveals that triglycerides are typical spectral features of actively growing non-necrotic tumors. Tosi et al. [102] showed that high-grade gliomas (GBM, anaplastic oligodendrogliomas) had prominent neovascularity and high-esterified cholesterol. No cholesterol esters were detected in healthy brain tissue or in low-grade and benign tumors. They suggest that highly esterified cholesterol could be a biochemical marker of malignancy. The presence of cholesteryl esters and triglycerides was found by Tugnoli et al. [103] to be correlated with the degree of vascular proliferation in high-grade brain tumors.
Microheterogeneity Using high-resolution magic-angle spinning (HRMAS) MRS from multiple specimens of a glioblastoma multiforme tumor, together with histopathology, Cheng et al. [104] noted, microheterogeneity is a routinely observed

112

neuropathologic characteristic in brain tumor pathology (p. 87). In addition, they found that lactate and mobile lipids linearly reflected percent tumor necrosis.

Phosphocholine to choline ratios Cheng et al. [104] also found that ratios of phosphocholine to choline correlated linearly with percentage of highly cellular malignant glioma. Similarly, in a study by Tzika et al. [105] the in vitro MRS obtained ratio of phosphocholine to choline was correlated with the percentage of cancerous tissue in brain tumors among children. Lipid to total creatine ratio correlated with the percentage of tumor necrosis in that study [105]. Myoinositol and glycine Barba et al. [45] performed in vitro MRS studies at 9.4T of hemangiopericytomas, in addition to their in vivo investigation. They were able thereby to confirm that in hemangiopericytomas the large peak at 3.56 ppm was due to high levels of myoinositol. Analysis of brain tumor biopsy specimens with HRMAS reveals that both myoinositol and glycine are elevated in astrocytomas, with decreasing grade, and that these metabolites are not present in meningiomas [32]. Kinoshita et al. [106] reported that glycine concentrations were particularly high in glioblastoma, using HRMAS on a total of 60 brain tumor specimens of various types. High inositol content was found in neurinomas. Metabolic Features of Meningiomas The authors of Ref. [106] also found that, in contradiction to some of the in vivo MRS findings, meningiomas did not have increased choline. The high alanine content of meningiomas was, however, confirmed by means of HRMAS. A high taurine content was found in medulloblastomas. Quantitative analysis by Czernicki et al. [107] revealed that only a four-times higher ratio of alanine to creatine plus phosphocreatine distinguished 11 low-grade meningiomas from 6 malignant gliomas. Acetate, Succinate, Valine & Leucine in Brain Abscesses vs Tumors In-vitro analysis by Grand et al. [108] demonstrated that acetate and succinate were present, respectively, in 12 and 8 of 13 brain abscesses, whereas these substances were much lower or barely detectible in necrotic brain tumors. Thus, these findings corroborate the observations from in vivo MRS, that the presence of acetate and succinate are suggestive of brain abscess rather than neoplasia. Furthermore, in vitro 2D J-resolved MRS showed much higher

113

concentrations of valine and leucine in the abscesses compared to tumor [108].

References
[1] D. Savitz, D. Trichopoulos, Brain Cancer, in: H-O. Adami, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, Oxford, 2002, p. 486-503. [2] V.A. Levin, S.A. Leibel, P.H. Gutin, Neoplasms of the Central Nervous System, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 2100-2160. [3] S.M. Sagar, M.A. Israel, Primary and metastatic tumors of the nervous system, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 2442-2452. [4] M. Blettner, H. Zeeb, A. Auvinen, et al., Mortality from cancer and other causes among male airline cockpit crew in Europe, Int. J. Cancer 106, 946-952 (2003). [5] E. Pukkala, R Aspholm, A Auvinen et al., Incidence of cancer among Nordic airline pilots over five decades: occupational cohort study, BMJ 325, 567 (2002). [6] M.K. Lim, Cosmic rays: are aircrew at risk? Occup. Environ. Med. 59, 428-432 (2002). [7] M.K. Hansson, L. Hardell, M. Kundi, M.O. Mattsson, Mobile telephones and cancer: is there really no evidence of an association? Int. J. Molecular Medicine 12, 67-72 (2003). [8] L. Hardell, K.H. Mild, M. Carlberg, Case-control study on the use of cellular and cordless phones and the risk for malignant brain tumors, Int. J. Radiat. Biology 78, 931-936 (2002). [9] P.D. Inskip, R.E. Tarone, E.E. Hatch, et al., Cellular-telephone use and brain tumors, N. Engl. J. Med. 344, 79-86 (2001). [10] M. Huncharek, B. Kupelnick, L. Wheeler, Dietary cured meat and the risk of adult glioma, J. Environ. Pathology, Toxicology & Oncology 22, 129-137 (2003). [11] B.S. Jhawar, C.S. Fuchs, G.A. Colditz, M.J. Stampfer, Sex steroid hormone exposures and risk for meningioma, J. Neurosurgery 99, 848-853 (2003). [12] D. Baris, G. Gridley, E, Ron, et al., Acromegaly and cancer risk: a cohort study in Sweden and Denmark, Cancer Causes & Control 13, 395-400 (2002). [13] J.C. Tamraz, C. Outin, M. Forjaz Secca, B. Soussi, MRI Principles of the Head, Skull Base and Spine A Clinical Approach, Springer, Paris, 2003. [14] A.E. Li, D.A. Bluemke, Magnetic Resonance Imaging in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 669-679. [15] J.B. Fiebach, P.D. Schellinger, A. Gass, et al., Stroke magnetic resonance imaging is accurate in hyperacute intracerebral hemorrhage A Multicenter study on the validity of stroke imaging, Stroke 35, 502-507 (2004). [16] D.H. Lowenstein, Seizures and epilepsy, in: E. Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 2354-2369.

114

[17] L.A. Brando, R.C. Domingues, MR Spectroscopy of the Brain, Lippincott Williams & Wilkins, Philadelphia, Pennsylvania, 2004. [18] N.H. Raskin, S.J. Peroutka, Headache, including migraine and cluster headache, in: E. Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 70-79. [19] B.E. Pollock, P.D. Brown, R.L. Foote, et al., Properly selected patients with multiple brain metastases may benefit from aggressive treatment of their intracranial disease, J. Neuro-Oncology 61, 73-80 (2003). [20] M. G. Pomper, Functional and metabolic imaging, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 679-689. [21] J.B. Fiebach, P.D. Schellinger, Stroke MRI, Springer, Darmstadt, 2003. [22] S. Gudowius, V. Engelbrecht, M. Messing-Junger et al., Diagnostic difficulties in childhood bilateral thalamic astrocytomas, Neuropediatrics 33, 331-335 (2002). [23] J. Walecki, E. Tarasow, B. Kubas, et al., Hydrogen-1 MR spectroscopy of the peritumoral zone in patients with cerebral glioma: assessment of the value of the method, Acta Radiol. 10, 145153 (2003). [24] O. Kadota, K. Kohno, S. Ohue, et al., Discrimination of brain abscess and cystic tumor by in vivo proton magnetic resonance spectroscopy, Neurol. Med. Chir. (Tokyo) 41. 121-126 (2001). [25] T.R. McKnight, M.H. van dem Bussche, D.B. Vigneron, et al., Histopathological validation of a three-dimensional magnetic resonance spectroscopy index as a predictor of tumor presence, J. Neurosurg. 97, 794-802 (2002). [26] D. Vigneron, A. Bollen, M. McDermott, et al., Three-dimensional magnetic resonance spectroscopic imaging of histologically confirmed brain tumors, Magn. Reson. Imaging 19, 89-101 (2001). [27] M.A. Brown, R.C. Semelka, MRI Basic Principles and Applications, 2nd Edition John Wiley & Sons, New York, 1999. [28] A. Romano, A. Bozzao, M. Bonamini et al., Diffusion-weighted MR imaging: clinical applications in neuroradiology, Radiol. Med. 106, 521-548 (2003). [29] K. Sartor, M. Hartmann, J. Fiebach, et al., Normal and abnormal water diffusion in the brain, Rofo. Fortschr. Geb. Rontgenstr. Neuen Bildgeb. Verfahr. 175, 1317-1329 (2003). [30] M. Nakaiso, M. Uno, M. Harada, et al., Brain abscess and glioblastoma identified by combined proton magnetic resonance spectroscopy and diffusion-weighted magnetic resonance imagingtwo case reports, Neurol. Med. Chir. (Tokyo) 42, 346-348 (2002) [31] N. Bulakbasi, M. Kocaoglu, F. Ors, et al., Combination of single-voxel proton MR spectroscopy and apparent diffusion coefficient calculation in the evaluation of common brain tumors, Am. J. Neuroradiol. 24, 255-233 (2003). [32] F.A. Howe, K.S. Opstad, 1H spectroscopy of brain tumors and masses, NMR Biomed. 16, 123131 (2003). [33] S.J. Nelson, Multivoxel magnetic resonance spectroscopy of brain tumors, Mol. Cancer Ther. 2, 497-507 (2003). [34] V. Govindaraju, K. Young, A.A. Maudsley, Proton NMR chemical shifts and coupling constants for brain metabolites, NMR Biomed. 13, 129-153 (2000).

115

[35] P.E. Sijens, M. Oudkerk, 1H chemical shift imaging characterization of human brain tumor and edema, Euro. Radiol. 12, 2056-2061(2002). [36] M. Utriainen, M. Komu, V. Vuorinen, et al., Evaluation of brain tumor metabolism with 11C choline PET and 1H-MRS, J. Neuroooncol. 62, 329-338 (2003). [37] D. Croteau, L. Scarpace, D. Hearshen et al., Correlation between magnetic resonance spectroscopy imaging and image-guided biopsies: Semiquantitative and qualitative histopathological analyses of patients with untreated glioma, Neurosurgery 49, 823-829 (2001). [38] C. Dowling, A.W. Bollen, S.M. Noworolski, et al., Preoperative proton MR spectroscopic imaging of brain tumors: correlation with histopathologic analysis of resection specimens, Am. J. Neuroradiol. 22, 604-612 (2001). [39] E. Saraf-Lavi, B.C. Bowen, P.M. Pattany, et al., Proton MR spectroscopy of gliomatosis cerebri: case report of elevated myoinositol with normal choline levels, Am. J. Neuroradiol. 24, 946-951 (2003). [40] T. Isobe, A. Matsumura, I. Anno, et al., Quantification of cerebral metabolites in glioma patients with proton MR spectroscopy using T2 relaxation time correction, Magn. Reson. Imaging 20, 343-349 (2002). [41] G. Fan, B. Sun, Z. Wu, et al., In vivo single-voxel proton MR spectroscopy in the differentiation of high-grade gliomas and solitary metastases, Clin. Radiol. 59, 77-85 (2004). [42] A.M. Saindane, S. Cha, M. Law, et al., Proton MR spectroscopy of tumefactive demyelinating lesions, Am. J. Neuroradiol. 23, 1378-1386 (2002). [43] R. Tarnawski, M. Sokol, P. Pieniazek, et al., 1H-MRS in vivo predicts the early treatment outcome of postoperative radiotherapy for malignant gliomas, Int. J. Radiat. Oncol. Biol. Phys. 52, 1271-6 (2002). [44] J.K. Smith, A. Londono, M. Castillo, L. Kwock, Proton magnetic resonance spectroscopy of brain-stem lesions, Neuroradiology 44, 825-829 (2002). [45] I. Barba, A. Moreno, I. Martinez-Perez, et al., Magnetic resonance spectroscopy of brain hemangiopericytomas: high myoinositol concentrations and discrimination from meningiomas, J. Neurosurg. 94, 55-60 (2001). [46] R.N. Sener, Astroblastoma: diffusion MRI, and proton MR spectroscopy, Comput. Med. Imaging Graph. 26, 187-191 (2002). [47] A.P. Lin, B.D. Ross, Short-echo time proton MR spectroscopy in the presence of gadolinium, J. Comput. Assist. Tomogr. 25, 705-712 (2001). [48] M. Murphy, D.J.O. McIntyre, H. Marsh, et al., Two-dimensional chemical shift imaging (2DCSI) for surveillance of low grade astrocytomas, MAGMA 15 (Suppl.1), 146-147 (2002). [49] E. R. Danielsen, B. Ross, Magnetic Resonance Spectroscopy Diagnosis of Neurological Diseases, Marcel Dekker, Inc., New York, 1999. [50] T. Kimura, K. Sako, T. Gotoh, et al., In vivo single-voxel proton MR spectroscopy in brain lesions with ring-like enhancement, NMR Biomed. 14, 339-349 (2001). [51] B. Kubas, E. Tarasow, W. Dzienis, et al., Magnetic resonance proton spectroscopy in neurooncologypreliminary report, Neurol. Neurochir. Pol. 35 (Suppl 5) 90-100 (2001). [52] W. Moller-Hartmann, S. Herminghaus, T. Krings, et al., Clinical application of proton magnetic resonance spectroscopy in the diagnosis of intracranial mass lesions, Neuroradiology 44, 371-381 (2002).

116

[53] D.J. Butteriss, A. Ismail, D.W. Ellison, D. Birchall, Use of serial proton magnetic resonance spectroscopy to differentiate low-grade glioma from tumefactive plaque in a patient with multiple sclerosis, Br. J. Radiol. 76, 662-665 (2003). [54] M. Murphy, A. Loosemore, A.G. Clifton, et al., The contribution of proton magnetic resonance spectroscopy (1H MRS) to clinical brain tumor diagnosis, Br. J. Neurosurg. 16, 329-334 (2002). [55] M. Castillo, J.K. Smith, L. Kwock, Correlation of myoinositol levels and grading of cerebral astrocytomas, Am. J. Neuroradiol. 21, 1645-1649 (2000). [56] X. Li, Y. Lu, A. Pirzkall, et al., Analysis of the spatial characteristics of metabolic abnormalities in newly diagnosed glioma patients, J. Magn. Reson. Imaging 16, 229-237 (2002). [57] L. Kwock, J.K. Smith, M. Castillo M, et al., Clinical applications of proton MR spectroscopy in oncology. Technol. Cancer Res. Treat. 1, 17-28 (2002). [58] A.A. Tzika, L.G. Astrakas, M.K. Zarifi, et al., Multiparametric MR assessment of pediatric brain tumors, Neuroradiology 45, 1-10 (2003). [59] M. Kaminogo, H. Ishimaru, M. Morikawa, et al., Diagnostic potential of short echo time MR spectroscopy of gliomas with single-voxel and point-resolved spatially localised proton spectroscopy of brain, Neuroradiology 43, 353-363 (2001). [60] D. Spielman, In vivo proton MR spectroscopy: basic principles and clinical applications, Section for Magnetic Radiation Technologists Educational Seminar 3, 19-37 (2000). [61] S.J. Nelson, T.R. McKnight, R.G. Henry, Characterization of untreated gliomas by magnetic resonance spectroscopic imaging, Neuroimaging Clin. N. Am. 12, 599-613 (2002). [62] M. Rijpkema, J. Schuuring, Y. van der Meulen, et al., Characterization of oligodendrogliomas using short echo time 1H MR spectroscopic imaging, NMR Biomed. 16, 12-18 (2003). [63] K. Zakrzewski, J. Kreisel, L. Polis, et al., Clinical application of proton magnetic resonance spectroscopy for differential diagnosis of pediatric posterior fossa tumors, Neurol. Neurochir. Pol. 35 (Suppl 5) 19-25 (2001). [64] S. Herminghaus, T. Dierks, U. Pilatus, et al., Determination of histopathological tumor grade in neuroepithelial brain tumors by using spectral pattern analysis of in vivo spectroscopic data, J. Neurosurg. 98, 74-81 (2003). [65] R. Nafe, S. Herminghaus, P. Raab, et al., Preoperative proton-MR spectroscopy of gliomas correlation with quantitative nuclear morphology in surgical specimen, J. Neurooncol. 63, 233-245 (2003). [66] H. Kugel, W. Heindel, R.I. Ernestus, et al., Human brain tumors: spectral patterns detected with localized 1H-MR spectroscopy, Radiology 183, 701-709 (1992). [67] P.S. Murphy, I.J. Rowland, L. Viviers, et al., Could assessment of glioma methylene lipid resonance by in vivo 1H-MRS be of clinical value? Br. J. Radiol. 76, 459-463 (2003). [68] M.C. Martinez-Bisbal, B. Celda-Muoz, L. Marti-Bonmati, et al., Contribution of MRS to the classification of high-grade gliomas. The predictive value of macromolecules. Rev. Neurol. 34, 309-313 (2002). [69] E.T.S. Smith, K. Tyler, K. Das, Can 2 minute duration single voxel proton MR spectroscopy be used successfully in routine clinical neuroradiological practice to grade cerebral gliomas? MAGMA 15 (Suppl 1), 146 (2002). [70a] C. Majs, J. Alonso, C. Aguilera, et al., Proton magnetic resonance spectroscopy (1H MRS) of human brain tumors: assessment of differences between tumor types and its applicability in brain tumor categorization, Eur. Radiol. 13, 582-591 (2003).

117

[70b] A. van den Boogaart, Quantitative data analysis of in vivo MRS data sets, Magn. Reson. Chem. 35, S146-S152 (1997). [71] D. Galanaud, O. Chinot, F. Nicoli, et al., Use of proton magnetic resonance spectroscopy of the brain to differentiate gliomatosis cerebri from low-grade glioma, J. Neurosurg. 98, 260-276 (2003). [72] D. Maintz, W. Heindel, H. Kugel, et al., Phosphorus-31 MR spectroscopy of normal adult human brain and brain tumors, NMR Biomed. 15, 18-27 (2002). [73] A. Del Sole, A. Falini, L. Ravasi, et al., Anatomical and biochemical investigation of primary brain tumors, Eur. J. Nucl. Med. 28, 1851-1872 (2001). [74] Y-D. Cho, G-H. Choi, S-P. Lee, J-K. Kim, 1H-MRS metabolic patterns for distinguishing between meningiomas and other brain tumors, Magn. Reson. Imaging 21, 663-672 (2003). [75] C. Majs, J. Alonso, C. Aguilera, et al., Adult primitive neuroectodermal tumor: proton MR spectroscopic findings with possible application for differential diagnosis, Radiology 225, 556-66 (2002). [76] C. Majs, J. Alonso, C. Aguilera, et al., Utility of proton MR spectroscopy in the diagnosis of radiologically atypical intracranial meningiomas, Neuroradiology 45, 129-136 (2003). [77] F.A. Howe, S.J. Barton, S.A. Cudlip, et al., Metabolic profiles of human brain tumors using quantitative in vivo 1H magnetic resonance spectroscopy, Magn. Reson. Med. 49, 223-232 (2003). L. Vanhamme, A. van den Boogart, S. van Huffel, Improved method for accurate and efficient quantification of MRS data with use of prior knowledge, J. Magn. Reson. 29, 35-43 (1997). [78] K.S. Opstad, S.W. Provencher, B.A. Bell, et al., Detection of elevated glutathione in meningiomas by quantitative in vivo 1H MRS, Magn. Reson. Med. 49, 632-637 (2003). S. W. Provencher, Automatic quantitation of localized in vivo 1H spectra with LCModel, NMR Biomed. 14, 260-264 (2001). [79] A.R. Tate, C. Majs, A. Moreno, et al., Automated classification of short echo time in in vivo 1H brain tumor spectra: a Multicenter Study, Magn. Reson. Med. 49, 29-36 (2003). [80] M. Law, S. Cha, E.A. Knopp, et al., High-grade gliomas and solitary metastases: differentiation by using perfusion and proton spectroscopic MR imaging, Radiology 222, 715-721 (2002). [81] W.A. Hall, A. Martin, H. Liu, C.L. Truwit, Improving diagnostic yield in brain biopsy: coupling spectroscopic targeting with real-time needle placement, J. Magn. Reson. Imaging 13, 12-15 (2001). [82] W.A. Hall, H. Liu, R.E. Maxwell RE, C.L. Truwit, Influence of 1.5 tesla intraoperative MR imaging on surgical decision-making, Acta Neurochir. Suppl. 85, 29-37 (2003). [83] H. Liu, W.A. Hall, A.J. Martin, C.L. Truwit, An efficient chemical shift-imaging scheme for magnetic resonance-guided neurosurgery, J. Magn. Reson. Imaging 14, 1-7 (2001). [84] A. Pirzkall, S.J. Nelson, T.R. McKnight, et al., Metabolic imaging of low-grade gliomas with three-dimensional magnetic resonance spectroscopy, Int. J. Radiat. Oncol. Biol. Phys. 53, 12541264 (2002). [85] S.J. Nelson, E. Graves, A. Pirzkall, et al., In vivo molecular imaging for planning RT of gliomas: an application of 1H MRSI, J. Magn. Reson. Imaging 16, 464-476 (2002). [86] M. Orszagh, C.B. Ostertag, Comment to [38], Neurosurgery 49,829 (2001). [87] E.E. Graves, S.J. Nelson, D.B. Vigneron, et al., Serial proton MR spectroscopic imaging of recurrent malignant gliomas after gamma knife radiosurgery, Am. J. Neuroradiol. 22, 613-624 (2001).

118

[88] R. Lodi, E. Setola, C. Tonon, et al., Gliomatosis cerebri: clinical, neurochemical and neuroradiological response to temozolomide administration, Magn. Reson. Imaging 21, 1003-1007 (2003). [89] A.A. Tzika, D. Zurakowski, T.Y. Poussaint, et al., Proton magnetic spectroscopic imaging of the childs brain: the response of tumors to treatment, Neuroradiology 43, 169-177 (2001). [90] J. Walecki, E. Tarasow, B. Kubas, et al., Hydrogen-1 MR spectroscopy of the peri-tumoral zone in patients with cerebral glioma: assessment of the value of the method, Acta Radiol. 10, 145153 (2003). [91] R.A. Kauppinen, Monitoring cytotoxic tumor treatment response by diffusion magnetic resonance imaging and proton spectroscopy, NMR Biomed. 15, 6-17 (2002). [92] F.W. Floeth, H.J. Wittsack, V. Engelbrecht, F. Weber, Comparative follow-up of enhancement phenomena with MRI and Proton MR Spectroscopic Imaging after intralesional immunotherapy in glioblastomaReport of two exceptional cases, Zentralbl. Neurochir. 63, 2328 (2002). [93] T. Kimura, K. Sako, Y. Tohyama, et al., Diagnosis and treatment of progressive spaceoccupying radiation necrosis following stereotactic radiosurgery for brain metastasis: value of proton magnetic resonance spectroscopy, Acta Neurchir. (Wien) 145, 557-564 (2003). [94] H. P. Schlemmer, P. Bachert, M. Henze, et al., Differentiation of radiation necrosis from tumor progression using proton magnetic resonance spectroscopy, Neuroradiology 44, 216-222 (2002). [95] J.D. Rabinov, P.L. Lee, F.G. Barker, et al., In vivo 3-T MR spectroscopy in the distinction of recurrent glioma versus radiation effects: Initial experience, Radiology 225, 871-879 (2002). [96] S.H. Ng, S.F. Ko, W.C. Chen, et al., Proton magnetic resonance spectroscopy of cerebral glioma after irradiation, Chang Gung Med. J. 24, 708-716 (2001). [97] F. Traber, W. Block, S. Flacke, et al., 1H-MR spectroscopy of brain tumors in the course of radiation therapy: use of fast spectroscopic imaging and single-voxel spectroscopy for diagnosing recurrence, Rofo. Forschr. Geb. Rontgenstr. Neuen Bildgeb. Verfahr. 174, 33-42 (2002). [98] Y.E. Kuznetsov, Z. Caramanos, S.B. Antel, et al., Proton magnetic resonance spectroscopic imaging can predict length of survival in patients with supratentorial gliomas, Neurosurgery 53, 565-576 (2003). [99] A.A. Tzika, L.G. Astrakas, M.K. Zarifi, et al., Spectroscopic and perfusion magnetic resonance imaging predictors of progression in pediatric brain tumors, Cancer 15, 1246-1256 (2004). [100] M.A. Thomas, L.N. Ryner, M.P. Mehta, et al., Localized 2D J-resolved 1H MR spectroscopy of human brain tumors in vivo, J. Magn. Reson. Imaging 6, 453-459 (1996). [101] D.P. Auer, C. Gssl, T. Schirmer, M. Czisch, Improved analysis of 1H-MR spectra in the presence of mobile lipids, Magn. Reson. Med. 46, 615-618 (2001). [102] M.R. Tosi, G. Bottura, P. Lucchi, et al., Cholesteryl esters in human malignant neoplasms, Int. J. Mol. Med. 11, 95-98 (2003). [103] V. Tugnoli, M.R. Tosi, A. Tinti, et al., Characterization of lipids from human brain tissues by multinuclear magnetic resonance spectroscopy, Biopolymers 62, 297-306 (2001). [104] L.L. Cheng, D.C. Anthony, A.R. Comite, et al., Quantification of microheterogeneity in glioblastoma multiforme with ex vivo high-resolution magic-angle spinning (HRMAS) proton magnetic resonance spectroscopy, Neuro-oncol. 2, 87-95 (2000). [105] A.A. Tzika, L.L. Cheng, L. Goumnerova, et al., Biochemical characterization of pediatric brain tumors by using in vivo and ex vivo magnetic resonance spectroscopy, J. Neurosurg. 96, 1023-1031 (2002).

119

[106] Y. Kinoshita, A. Yokota, Absolute concentrations of metabolites in the human brain tumors using in vitro proton magnetic resonance spectroscopy, NMR Biomed. 10, 2-12 (1997). [107] Z. Czernicki, D. Horsztynski W. Jankowski, et al., Malignancy of brain tumors evaluated by proton magnetic resonance spectroscopy (1H-MRS) in vitro, Acta Neurochir. Suppl. 76, 17-20 (2000). [108] S. Grand, G. Passaro, A. Ziegler, et al., Necrotic tumor versus brain abscess: Importance of amino acids detected at 1H MR spectroscopyInitial results, Radiology 213, 785-793 (1999).

121

Chapter 9

Prostate Cancer
_______________________________________________________________________________

Almost all cancers of the prostate are adenocarcinomas, and about 75% arise from the peripheral zone [1, 2].

9.1

Overview of epidemiological & clinical aspects

Incidence and prevalence/morbidity and mortality Worldwide, prostate cancer is third in incidence among men and the sixth cause of cancer deaths [2]. Among men who died in the 8th decade of life and were autopsied, over 90% had hyperplastic changes in the prostate and malignant changes were seen in over 70% [3]. Five-year survival among European men with prostate cancer diagnosed between 1985 and 1989 was between 37% and 72%. The mortality to incidence ratios in the U.S. (among whites) was 0.16, compared to 0.6 in Denmark. A high survival rate, and as a corollary a low mortality-to-incidence ratio, may be due primarily to detection of many cancers unlikely to progress to advanced lethal disease (p. 401) [1]. The number of detected cases has increased dramatically in the 1990s based on the widespread testing with prostate specific antigen (PSA) [3]. The highest rates are reported in the U.S. and Western Europe and lowest in the Asian countries. However, these findings must be interpreted in the light of diagnostic intensity and screening behavior. In other words, incidence rates in some countries, the United States being a prime example, reflect the sum of clinical disease and latent disease, but in other countries only clinical disease (p. 401) [1]. Nevertheless, even before PSA became available, there were marked geographic differences in incidence.

122

Migration studies reveal increased risk among native-born Japanese and Polish people who moved to the U.S., although screening practices could also affect these findings [1]. In the U.S. African Americans are at highest risk, and are frequently diagnosed at a late stage. This late detection is noted to be inversely correlated with income and education [2]. Etiology/risk factors Age This is the best-documented risk factor, though it is less apparent in Asia. In western countries prostate cancer rates show the steepest agedependent rise of any cancer [1]. Inherited Susceptibility There seems to be a heritable component to risk of prostate cancer. This is particularly apparent among those who develop prostate cancer at a younger age. Some evidence exists for autosomal dominant susceptibility. Risk of prostate cancer is reportedly increased among carriers of germline mutations of BRCA 1 or 2 [1]. Exposures related to lifestyle
Alcohol Very heavy alcohol consumption is associated with increased risk of prostate cancer [1]. Diet Animal fat consumption Consistent evidence of a positive association. The increased fat intake may explain, at least in part, the increased incidence of prostate cancer among Japanese immigrants to the U.S. [1, 2]. Lycopene intake From cooked or processed tomato products. There is longitudinal evidence of a protective effect, although case-control studies are less convincing [1]. Selenium Shown to inhibit growth and stimulate apoptosis in human prostate cancer cell lines. The role of selenium has been difficult to assess epidemiologically, with likely misclassification. Clinical trials are on going for secondary prevention, for those undergoing watchful waiting, among high-risk populations and among the general population [1, 4, 5]. Vitamin E Antioxidant, which can reduce DNA damage and inhibit malignant transformation of cells and stimulates the immune system [1]. Evidence from a Finnish randomized, placebo-controlled intervention trial among 29 133 male smokers aged 50-69 at 5-8 years, that with Vitamin E supplementation there was a significant (32%) decrease in prostate cancer incidence and 41%

123

decrease in prostate cancer mortality, with an elevated prostate cancer incidence and mortality among those receiving beta-carotene [6]. However, other studies have not supported a protective role for Vitamin E [1].

Occupation Some data, though not entirely consistent, suggest an increased risk with farming [1]. Recent findings from the Central California Cancer Registry indicate that among unionized farm workers (mainly Hispanic) high levels of exposure to organochlorine and organophosphate pesticides, fumigants and triazine herbicides were associated with increased risk of prostate cancer compared to those with lower levels of exposure [7]. A study of a representative sample of 58 678 men in Canada using a job exposure matrix revealed a significantly increased relative risk for prostate cancer among those exposed to calcium carbonate or metal dust [8]. Clinical Presentation and Approach Carcinoma of the prostate is usually asymptomatic in the early stages. In contrast, benign prostatic hypertrophy (BPH) encroaches on the urethra leading to symptoms of outlet obstruction: hesitancy, intermittent voiding, decreased urinary stream, incomplete emptying, and post-voiding leakage. These symptoms are reflected in the American Urological Association Index comprised of seven questions on a scale from 0 to 5 (incomplete voiding, frequent urination, intermittent voiding, urgency, weak urine stream, strain to begin urinating, arising from sleep to urinate). The symptoms of BPH are somewhat distinct from irritative symptoms (frequency, dysuria and urgency). Later stage and more unusual manifestations of prostate cancer include: bone pain, myelophthisic disorders, disseminated intravascular coagulation and spinal cord compression. All of these are much less common since PSA has facilitated early diagnosis [3]. Diagnosis Digital Rectal Exam (DRE) Prostate cancer is typically hard, nodular and irregular, but this may also be seen when BPH is accompanied by fibrosis. Cancer of the prostate often begins in the posterior surfaces of the lateral lobes; in this location it is easily palpable [3]. The digital rectal exam detects fewer prostate malignancies than does PSA, and when detected these are usually at a later stage.

124

However, DRE sometimes detects cancers in the face of normal PSA [2]. Prostate Specific Antigen PSA is a prostate-specific serum protease produced by cancerous, as well as non-cancerous, epithelial cells. Normal levels are below 4ng/mL. In the range of 4 10ng/mL about 30% of men will have a prostatic malignancy and approximately 50% will have cancer if PSA is above 10 ng/mL [3]. Most cancers detected with PSA are clinically significant, but clinically confined [2]. However, among 9 459 men aged 62 to 91 enrolled in the Prostate Cancer Prevention Trial who received placebo and annual PSA and DRE, 2 950 never had a PSA > 4 ng /mL or an abnormal digital rectal examination. After 7 years, biopsy was performed and prostate cancer was found in 449 (15.2%) of these men, with 67 of these 449 cancers (14%) having a Gleason score 7. The authors [9] conclude: biopsy-detected prostate cancer, including high-grade cancers, is not rare among men with PSA levels of 4 ng per milliliter or lesslevels generally thought to be in the normal range (p. 2239). They point out that this study is not subject to verification bias since only men who underwent an end-of-study biopsy are included in the analysis. This study raises many questions about screening with PSA, and the threshold level for biopsyespecially regarding over-treatment of clinically unimportant disease if the threshold is lowered versus the risk of under-treating potentially clinically important disease if the PSA level remains at 4.0 ng/mL.
Non-malignant causes of alterations in PSA The most frequent non-malignant cause of increased PSA is BPH. The PSA can also be increased temporarily with prostatitis, endoscopic urethral manipulation, biopsy and to a lesser extent with ejaculation. PSA will be lowered by luteinizing hormone releasing hormone (LHRH) agonists and antagonists, and by the 5-reductase inhibitor finasteride, which inhibits the conversion of testosterone to dihydrotestosterone1.

Refinements of PSA measurements

Prostate Specific Antigen Density (PSAD)

A correction for BPH (measured on ultrasound) can be made, yielding the prostate-specific antigen density, with values below 0.1 consistent with BPH and above 0.15 suggestive of malignancy. With age, the PSAD increases [3].
1 Finasteride is being tested as a preventive agent for prostate cancer. Dihydrotestosterone is a more potent stimulator of prostate cell proliferation than is testosterone. Animal studies suggest that finasteride may inhibit the induction of prostate cancer. The initial clinical results include both a protective role and increased risk of higher-grade disease.

125

Prostate Specific Antigen Velocity

This assesses the rate of change in PSA prior to the establishment of a cancer diagnosis. If this is above 0.75 ng/mL per year, malignancy is suspected.
Percentage of free PSA

This may be used to improve diagnostic accuracy and to predict aggressiveness [2].

Transurethral Ultrasonography (TRUS) Most prostate cancer is hypoechoic on TRUS. Those smaller than 5-7 mm that are well differentiated and that are located within the transition zone are difficult to distinguish from normal prostate tissue. TRUS aids in guiding biopsy sampling. Biopsy has a sensitivity of approximately 80% [3]. Prostate Biopsy Biopsy is indicated for elevated PSA and/or DRE, and is best performed with TRUS guidance. However, TRUS as well as DRE lack sensitivity and specificity in this respect. Since at least 30% of cancers originate in the transitional zone, many consider that this region should be biopsied [2]. CT, PET and other non-MRI based imaging modalities CT has been used to exclude lymph node metastases [2]. FDG-PET is found to be of limited value for prostate cancer due to its generally low metabolic activity [10]. Radionuclide bone scans are used for detecting metastatic disease. These show high sensitivity, but low specificity for prostatic metastasis, with false positive findings due to trauma, degenerative disease or Pagets disease [2]. Differential diagnosis The symptoms seen with prostate cancer can also occur with acute prostatitis, granulomatous prostatitis, and prostatic calculi. The physical findings can also be seen with BPH if fibrosis is present. Metastases are rare to prostate, but may occur from bladder or colonic lesions [3] (see also sub-section on PSA). Classification and grading Pathology Prostatic intraepithelial neoplasia (PIN) is considered a precursor lesion, which does not always progress to invasive cancer.

126

Poorly differentiated tumors have poorer prognosis. The worst differentiated site frequently determines biological behavior [3].
Gleason Grade Scores range from 1 to 5 based on the two most predominant histological patternsThese are summed to yield the Gleason sum [2]. Gleason Sums 2 to 4 = well-differentiated disease 5 to 7 = moderately differentiated disease 8 to 10 = poorly differentiated disease

Staging (TNM)2[2,3]
TIS=Carcinoma in situ (PIN) T1a = Nonpalpable, 5% resected tissue contains cancer T1b = Nonpalpable, > 5% resected tissue contains cancer T1c = Nonpalpable, detected on the basis of increased PSA T2a = Palpable, of 1 lobe T2b= Palpable, > of 1 lobe, but not 2 lobes T2c= Palpable, involves both lobes T3a =Palpable, unilateral extracapsular extension T3b =Palpable, bilateral extracapsular extension T3c =Invasion of seminal vesicles M = Distant metastases

Most men are now diagnosed at stage T1c or T2. Extracapsular extension does not necessarily mean that the cancer cannot be cured [2].

Treatment and prognosis

No cancer diagnosis: Screening The American Cancer Society and the American Urological Association recommend yearly DRE and PSA determination for men aged 50-79. If there is a 1st degree relative with prostate cancer or if the patient is African American, screening is recommended beginning at age 45.
2

TNM is the acronym for tumor, lymph nodes and metastases

127

Screening is recommended despite the lack of prospective, randomized controlled evidence demonstrating the benefit. However, this approach has led to an increased detection of localized tumors, decreased frequency of nodal spread and of late stage detection. On the other hand, over-detection and over-treatment can entail morbidity and even mortality [3]. Elevated PSA and no malignancy on biopsy Continued monitoring in such cases is recommended, sometimes with repeated biopsy. Active research is on going concerning how to prevent progression from PIN in the transition zone, to invasive cancer. Clinically localized disease Comparisons among these approaches are limited by the lack of prospective comparative trials, referral biases, and differences in the endpoints evaluated (p. 613) [3].
Conservative management (watchful waiting) Case selection criteria are being developed. Not recommended for those with high-grade disease or over 10-year life expectancy. Note the difficulty of monitoring progression so that the window of curability is not lost{since} the disease is often multifocal the biopsy on which the decision to defer therapy is made may not represent accurately the malignant potential of a second unidentified cancer (p. 613) [3]. It is difficult to evaluate the watchful waiting approach in a completely reliable manner because of selection. Radical prostatectomy Aims to remove all prostate tissue with a clear margin of resection, but to preserve external sphincter to preserve continence. Mortality < 0.4%, rare complications, which include rectal injury, deep vein thrombosis and embolic events. Urinary incontinence occurs in 5 to 10% by physician report (higher if reported by a third party). There is a higher risk in older patients. Incontinence usually occurs immediately post-operatively, but most patients recover urinary continence within 1 year if nerves have been preserved. Fecal incontinence is termed a rare complication. Impotency usually occurs after the procedure, but recovers (not to baseline though) in 6-12 months. Younger age and maintenance of neurovascular bundles are associated with better recovery. With sparing of the neurovascular bundle, pre-operative potency is recovered in up to 68% of patients [2,3]. Within 4 weeks post-operatively, PSA should become undetectable. Otherwise, recurrent or persistent disease is likely. Pathologic stage is the strongest predictor of recurrence [3].

128

Radiation therapy Considered a curative modality, with improved results over the last 3 decades.
External beam therapy For stages T1 and T2a similar results have been obtained as with radical prostatectomy. Local recurrence rates are inversely related to total dose. Complications are directly related to total dose. Early complications include those involving the rectum: diarrhea, rectal pain, bleeding, and those involving the bladder: urinary frequency, bleeding and pain. Late complications (3-6 months after therapy) include those involving the rectum: diarrhea, bleeding, perforation, and fistulae, and those involving the bladder: cystitis, bleeding and diminished capacity. There are fewer complications with 3D Conformal RT, i.e. with RT that conforms to the anatomic boundaries of the tumor [2]. Al-Abany [11] reports that risk of fecal leakage is significantly correlated with the dose-volume histogram of the anal-sphincter region. He underscores the importance of careful monitoring to reduce the amount of radiation to the analsphincter region. Interstitial therapy Interstitial brachytherapy provides intensive irradiation to the prostate with minimum to the surrounding tissue. Usually well tolerated and associated with a good prognosis, although patients with favorable features are usually chosen for this treatment modality. Incontinence occurs in 2-4% of patients [3].

Rising PSA Need to determine whether this is local persistence/recurrence, where additional therapy could be curative versus micrometastatic disease. If recurrence occurs after RT, salvage prostatectomy may be considered if no metastases are present and the cancer is amenable to surgery. Metastatic disease This is most frequently treated with androgen blockade, with gonadotropin-releasing hormone analogues, leading to rise in luteinizing hormone and follicle stimulating hormone, which downregulates receptors (although there is first a rise in testosterone). Also, diethylstilbestrol has been used, with many cardiovascular complications. Progesterone or non-steroidal anti-androgens, e.g. flutamide, have been used to block the binding of androgens to the receptor; gynecomastia, fatigue and increased liver enzymes are among the side effects [3].

129

9.2

MR-assisted prostate cancer diagnostics

Ornstein and Kang [12] note the limited specificity of PSAup to 75% of patients undergoing biopsy for elevated PSA do not have cancer of the prostate. Moreover, sextant biopsies of the prostate often miss the cancer (over 20% are missed). To decrease the number of falsenegative biopsies, obtaining at least 12 cores is recommended during the biopsy session. This procedure is considered safe and yields a marginal improvement in diagnosing prostate cancer. The authors stress the important emerging role of imaging to improve prostate cancer diagnostics. 9.2.1 MRI for prostate cancer diagnosis and staging

Technical Considerations An endorectal coil alone or together with anterior phased array coils are usually used for MR imaging of the prostate and surrounding structures [13]. An external pelvic phased array surface coil can be used with comparable accuracy to an endorectal surface coil for patients with rectal diseases or who otherwise cannot tolerate an endorectal procedure [14]. The endorectal coil, however, is considered to provide better views of the prostate and the surrounding region, compared to images obtained using a body coil [15]. The volume of air introduced into the balloon is of critical importanceif underinflated poor SNR is obtained for the anterior aspect of the prostate. Overinflation can displace the coil superiorly past the prostate, also compromising SNR. The use of a pelvic phased-array coil plus an endorectal coil provides high resolution of the entire gland and surrounding region, and is the preferred coil combination. Glucagon administration immediately prior to signal acquisition reduces bowel peristalsis, thereby improving image quality. After biopsy, hemorrhage is often present; this can compromise interpretation. Macilquham et al. [15] therefore recommend waiting at least 3 weeks after biopsy. T1 weighting has also been used to help identify an area of hemorrhage. Appearance of prostate cancer on T2 weighted MRI With T2 weighting, prostate cancer typically appears as low signal intensity. There are, however, wide ranges of diagnostic accuracy of MRI for prostate cancer. MRI shows low sensitivity for central zone cancers, and poorly distinguishes benign from malignant lesions [13, 16, 17]. Prostate cancer staging with MRI Capsular penetration may be seen with MRI as gross tumor extending into the surrounding fat or as irregularity or thickening of the capsule. Asymmetric enlargement of the neurovascular bundle on MRI suggests tumor invasion of that structure. Thickened or low T2 intensity of the seminal vesicles may be seen if these are involved. Compared to CT, superior anatomic detail is provided by MRI, which is the preferred imaging modality for staging of

130

prostate cancer. However, the diagnostic accuracy for staging with MRI has reportedly ranged from 51 to 92% [13]. Contrast enhancement - Perfusion imaging Gadolinium enhancement is usually considered of marginal value in MRI of the prostate (p. 39) [15]. However, there is some suggestion that CEperfusion imaging may help evaluate prostate tumor vascularity, depiction of capsular penetration and seminal vesicle infiltrationall of this is still under investigation [15].

9.2.2

MRSI in the primary diagnosis of prostate cancer

Coakley et al. [18] consider that MRSI as well as MRI have a limited role in prostate cancer diagnosis, but may be helpful for patients with a high index of suspicion and negative initial biopsy (p. S69). On the other hand, Wu et al. [19] point out: regions of absent or low citrate concentration in the prostate can be visualized at a resolution of a few mm. This new advancement provides a tool for early diagnosis and screening (p. 1577). Similarly, Macilquham et al. [15] note the excellent spatial localization for MRI, and cite pioneering studies {showing} the potential of combining MRI and MRS to determine the presence of prostate cancer more accurately and to estimate localization, extent, and severity of disease (p. 37).
The importance of citrate as a normal metabolite of the prostate The normal prostate accumulates and secretes very high amounts of citrate produced by non-malignant prostate epithelial cells. Citrate is lowered with malignancy, since prostate cancer cells oxidize citrate [20]. A correlation has been found between T2 relaxation time of water and citrate concentration. Namely, in malignancy the citrate concentration decreases and T2 water values decrease [17]. Empirical Studies In a retrospective study by Dhingsa et al. [21] of 37 patients with biopsyproven prostate cancer, reader awareness of the clinical data (DRE, PSA and histopathology) was found to significantly increase the detection of prostate cancer nodules using MRI and MRSI3, but this also increased false positive readings. MRI was overlaid with voxels containing at least 75% peripheral zone tissue and had an SNR > 5. The criterion for cancer was (choline + creatine) to citrate ratio that was 3 Sd > normal mean values.

Dhingsa et al. [21] used 1.5T, MRSI, PRESS, TR/TE=1000/130 ms, calculated (choline + creatine) to citrate ratios.

131

Yuen et al [22] applied MRI and MRSI4 among 24 patients with 1 to 3 negative TRUS-guided prostatic biopsies for persistently elevated PSA and/or abnormal DRE. MRI criteria An area was considered suspicious on MRI if T2 intensity was discrete and homogeneously low and was not an area of hemorrhage as detected on T1 weighted scans. An area was considered equivocal on MRI if there was low, heterogeneous signal on T2 weighted images. MRSI criteria (Choline + creatine) to citrate ratio < 1, considered normal, (Choline + creatine) to citrate ratio = 1 considered equivocal, (Choline + creatine) to citrate ratio > 1 considered abnormal. Prostate cancer was detected in 7 of the 24 patients (29.2%). In 2 of the 7 patients the cancers would have been missed if additional biopsies had not been directed to the abnormal areas detected by MRI and MRSI. Six of 21 MRSI labeled suspicious cores and 6 of 12 MRI labeled suspicious cores were positive for cancer on histology. Adding MRSI to MRI increased sensitivity and negative predictive value to 100%. Comments and limitations of this study 15 of 21 MRSI labeled suspicious cores and 6 of 12 MRI labeled suspicious cores were negative for cancer on histology. The authors [22] note that different investigators use different spectroscopic criteria for prostate cancer. In this study the criteria were qualitative. Among the 7 patients with prostate cancer, 6 had discordant findings on MRI and MRSInamely one was positive and the other negative, or vice versa. Besides small size of the study, other problems include the fact that phase correction was entirely automatic without the possibility for manual correction, it was not possible to shift the voxel to avoid central gland (CG) and extraprostatic fat, manual frequency shifting was also not possible. Conclusion This can be considered a pilot study, but one that suggests the possible role for MRI and MRSI for patients with persistently abnormal PSA.

MRSI for distinguishing BPH from prostate cancer Not only does a normal-appearing MRI not exclude malignancy with complete confidence, but choline can also be increased with BPH, due to high cell proliferation [15]. BPH as well as cancer can show metabolic heterogeneity [23].
Citrate in BPH versus prostate cancer Citrate helps distinguish glandular BPH from cancer, since in the former there is high citrate production, whereas with malignant transformation, citrate is oxidized rather than synthesized [24]. However, stromal BPH can have citrate levels similar to those of cancerous tissues, as is known from in vitro studies [23] (see sub-section on metabolic features of various zones of the prostate).
4

Yuen et al. [22] used 1.5T, MRSI, PRESS, TR=1000 ms, TE=130 ms, automatic post-processing with commercially available software.

132

Empirical Studies In a small MRS5 study by Liney et al. [25] citrate concentrations were estimated in normal peripheral zone (PZ) of the prostate, in BPH and in patients with newly diagnosed prostate cancer, with PSA >4 ng/ml, Gleason 3 to 5 from TRUS guided biopsy. As seen from Table 9.1, the mean citrate concentration was highest in BPH and lowest in prostate cancer. Besides MRS, these authors also applied dynamic contrast enhanced MRI. They calculated the Enhancement Factor (EF)6 and time to reach maximum enhancement (Tmax). Tmax was significantly correlated with citrate concentration, r = 0.712, p = 0.001. The authors note that CE MRI with faster imaging sequences can be helpful, especially with poorly differentiated tumors. Some controversy exists about the best MRI parameters to use; in this study [25] it was the Tmax showing much faster enhancement in areas of cancer. This may reflect neovascularization. On the other hand, EF variables showed substantial overlap among the various regions.

Table 9.1 Citrate in BPH, Normal PZ, and Prostate Cancer

(From data of Liney et al. [25]) Citrate Concentration

Clinical Groups BPH

(9 samples)

(mol/g wet weight) Mean Sd

65.7 13.2

Normal PZ
(4 samples)

42.4 19.5

Cancer
(9 samples)

5.9 5.8

Garca-Segura et al. [24] compared MRS7 findings in 10 patients with BPH and 10 with prostate cancer (locally advanced T3 or T4) or disseminated palpable newly diagnosed untreated disease.

5 Liney et al. [25] used 1.5T, 0.4 to 4.4 ml voxel volume, careful exclusion of fat, selected areas of suspected tumor, STEAM, TR=1000 ms, TE=30ms, citrate peak areas were obtained from manually phased spectra by integration over a fixed frequency range, which excluded the area of the rapidly modulating outer peak of the citrate spectrum (p. 40), corrected for J coupling effects. 6 EF = ln [(Smax-S0)/( Smax-St)], where Smax is maximum possible signal under fully relaxed conditions, St is the signal intensity at time t, and S0 is the average pre-contrast signal intensity. 7 Garcia-Segura et al. [24] used 1.5T, single voxel, STEAM, TR=1500ms, TE=50 ms, peak areas using least square fitting, and then ratios.

133

The patients with BPH were diagnosed on the based of DRE, PSA, TRUS and subsequent surgical excision. All adenocarcinoma were confirmed by TRUS guided biopsy. Patients were selected by having large, homogeneous lesions that over-encompassed the VOI, i.e. well-defined pathologic volumes > 4 mL. None were receiving therapy that could alter the metabolic activity of the prostate. Metabolite assignments: citrate 2.6-2.7 ppm, creatine plus phosphocreatine 3.0 ppm, choline containing compounds 3.2 ppm, myoinositol 3.6 ppm.

Table 9.2 Area ratios in BPH and advanced prostate cancer

(From data of Garca-Segura et al. [24])

Citrate/Choline
Mean Sd (Range)

Creatine/Choline
Mean Sd (Range)

Creatine/Myoinositol
Mean Sd (Range)

10 Patients with BPH

3.01 0.86 (2.09 5.04) ***

0.49 0.07 (0.35 0.59) *** 0.21 0.14 (0 0.40)

1.41 0.48 (1.02 2.77) *** 0.63 0.31# (0 0.96)

10 Patients with Prostate Cancer

0.39 0.28 (0 0.78)

*** p < 0.001 Kruskall-Wallace test # Myoinositol was not obtained in 1 patient

Findings The combination of citrate/choline and creatine/myoinositol ratios >1 distinguished the two groups. All the patients with prostate cancer in this study had a large choline peak. Myoinositol was the third highest peak in the patients with prostate cancer: an additional distinctive feature of this pathology (p. 759). Interpretation Myoinositol is generally not detected in clinical, in vivo studies of the prostate (see in vitro subsection). Phosphatidyl inositol 4,5-biphosphate (PIP2) is a membrane component that is hydrolysed by phospholipase, producing inositoltriphosphate and diacylglycerol as second messengers. The end product is myoinositol, which is re-utilised for PIP2 synthesis. Hydrolysis of PIP2 has

134

been found to correlate with some oncogene products. Thus, myoinositol might act as an indirect sign of oncogenic conditions (p. 763). The mobile lipid resonance was centered at 1.3 ppm. This is not considered to be contamination because the lesion over-encompassed the voxel. This was present in 8 of the 10 patients with prostate cancer. These authors note that unlike the spectrum of the patients with BPH, which, in all cases, was dominated by the citrate peak at 2.6 2.7 ppm, the most intense resonance in the patients with prostate cancer was at 1.3 to 1.4 ppm. This is suggested to indicate that in these patients with advanced disease (unlike those with small cancerous lesions), necrosis was a prominent feature. Caveats and limitations While it is true that there is no actual overlap, for creatine/myoinositol there nearly was in 4 cases. MRS for prostate cancer has rather limited sensitivity in the analysis of smaller lesions{as is the usual presentation of} prostate cancers, together with their tendency to have a spectroscopically inconvenient peripheral location (p. 763). The authors [24] underscore the need for a quantitative approach with absolute metabolite concentrations rather than ratios.

Metabolic features of different zones of the prostate There are zonal differences, with the periphery being mainly involved in producing citrate and therefore containing the highest concentrations. However, occasionally overlap is found between the lowest citrate levels in BPH and the highest citrate levels in prostate cancer. Stromal tissue normally has low citrate levels [20]. BPH mainly affects the CG whereas adenocarcinoma is seen in the peripheral zone in about 70% of cases, but can also invade or arise from the central portion [17].
Empirical Studies

An MRSI study8 by Liney et al. [17] included 16 patients with newly diagnosed prostate cancer, all of whom had elevated PSA, a TRUS-guided confirmatory biopsy 4-6 weeks prior to the MR examination, and Gleason score of 3 to 5. The citrate concentration was compared in the normal CG and PZ regions and regions of prostate cancer as shown in Table 9.3 For all regions taken together, r = 0.903, p < 0.0001 between T2 relaxation time of water and citrate. T2 relaxation time of water is expected to be short with cancerous tissue. The PZ was often compressed by the CG.

Liney et al. [17] used 1.5T, 10 mm voxel (in 1 case 15 mm), careful exclusion of fat, selected areas of suspected tumor, STEAM, TR=1000 ms, TE=30ms, citrate peak areas were obtained from manually phased spectra by integration over a fixed frequency range, which excluded the area of the rapidly modulating outer peak of the citrate spectrum (p. 1178), corrected for J coupling effects, in peripheral zone or in CG depending upon pathology.

135

Table 9.3 Citrate in Normal PZ, CG and Prostate Cancer

(From data of Liney et al. [17]) Citrate Concentration Clinical Sample Normal CG Region
(8 samples)
(mol/g wet weight) Mean Sd

64.5 13.6 42.4 19.5 5.9 5.8

Normal PZ Region
(4 samples)

Tumor Region
(9 samples)

The transition zone

Zakian et al. [23] examined the metabolic pattern of cancers of the prostatic transition zone (TZ). Prior to radical prostatectomy endorectal MRI and MRSI9 were performed in 40 patients with prostate cancer in the TZ identified with step-section pathologic analysis. Among these were 16 patients with tumors of largest diameter 1 cm who were included in this study with comparisons to normal TZ tissue. The results are summarized in Table 9.4.

Table 9.4 Metabolite ratios in normal and cancerous transition zone of the prostate (From data of Zakian et al. [23])
(Cho+Cre)/Citrate (Mean Sd) Cancerous TZ 1.74 1.35 ** Normal TZ 0.63 0.20 Choline/Creatine (Mean Sd) 3.01 1.61 ** 1.70 0.89 Choline/Citrate (Mean Sd) 1.28 1.16 ** 0.35 0.11

** p < 0.01, Wilcoxon paired test

9 Zakian et al. [23] used 1.5T, MRSI, PRESS, TR/TE=1000/130 ms, 2Hz Lorentzian spectral apodization, 4D Fourier transform, automated frequency, phase and baseline correction, zero filling to 3.1 mm resolution in the superior-inferior dimensions. Peak areas were calculated by numerical integration. Any metabolite with SNR < 1 was assigned the noise value.

136

Nine of the 16 patients had at least one voxel within the tumor showing only choline as a detectable peak. This was not the case for any of the control (i.e. BPH) voxels. The McNemar test shows expression of only choline as a metabolite was significantly more common in the tumor (p = 0.008). However, the percentage of voxels showing no metabolites whatsoever did not differ between tumor and controls. Caveats and technical issues: There was substantial overlap between the metabolite ratios in the tumor versus control regions. This overlap may be related to BPH, which has variable metabolite ratios. As mentioned earlier, both BPH and cancer can show metabolic heterogeneity. Also, stromal BPH can have citrate levels similar to those of cancerous tissues, as is known from in vitro studies. Tumors containing non-cancerous epithelial cells have substantial amounts of citrate. The anterior portion of the prostate is the most distant from the endorectal probe, and has lower SNR. The TZ is close to the urethra, which can contain glycerophosphocholine in seminal fluid.

The authors from Ref. [23] underscore the importance of assessing the TZ, noting that the PZ is usually targeted in biopsies. As a consequence, many cases occur in which there is a rising PSA, negative biopsy but cancer is actually present in the TZ. It has even been suggested that patients with rising or increased PSA levels and multiple negative biopsies, should have the TZ targeted for examination. Moreover, prostate cancer, which is limited to the TZ, tends to be more confined, with lower Gleason scores and higher biochemical cure rates. Patients in this category might also be candidates for watchful waiting. They conclude: although a trend towards elevation of choline and reduction in or lack of citrate was observed in TZ tumors, where they were compared with BPH, the broad range of metabolite ratios observed precludes the use of a single ratio to differentiate TZ cancer from benign TZ tissue. A prospective study is now needed to ascertain the value of combined MR imaging/MR spectroscopic imaging in the detection of TZ tumors (p. 246).

9.3

MRI and MRSI for treatment planning

Assessment of Tumor Extent/Staging MRSI has been used to assess tumor location and extent. However, the accuracy of volumetric localization is limited for tumors less than 0.5cc. Staging with MRSI is reported to improve the accuracy of MRI, and this is of prognostic significance for patients with moderate or high-risk cancers. MRSI can improve assessment of extracapsular

137

spread of prostate cancer and its aggressiveness. A decrease in citrate and increased choline show a strong linear correlation with cancer aggressiveness, as gauged by the Gleason grade [18, 26].
Empirical studies

Coakley et al. [27] performed endorectal MR and 3D MRSI10 in 37 patients prior to radical prostatectomy. They defined cancer as a (choline + citrate) to citrate ratio exceeding 3 Sd above normal mean values. A positive correlation was found between MRI and 3D MRSI and both with histopathologic volume of tumor. Scheidler et al. [16] performed MRI and 3D MRSI11 in 53 patients with biopsy-proven prostate cancer and subsequent radical prostatectomy and stepsection histopathologic examination. Possible cancer was defined as 2 Sd > normal for (choline + creatine) to citrate ratio, which was when that ratio exceeded 0.75. They performed a blinded evaluation using the two methods; 3D MRSI had a significantly higher specificity, but lower sensitivity than MRI. High specificity (up to 90%) was obtained with combined MRI and 3D MRSI, with high sensitivity (up to 95%) when either method alone indicated a positive result. The authors conclude: the addition of 3D MR spectroscopic imaging to MR imaging provides better detection and localization of prostate cancer in a sextant of the prostate than does use of MR imaging alone (p.473).
Extracapsular extension

Yu et al. [28] demonstrated that the addition of 3D MRSI12 to MRI improves the accuracy of diagnosing extra-capsular extension of prostate cancer, which is a key factor with respect to prognosis and choice of treatment.

Selection of Therapeutic Modality

To improve therapeutic selection: (e.g. watchful waiting versus active therapy)

Kuranewicz et al. [29] present a case in which PSA and biopsy suggested that watchful waiting would be appropriate, whereas the finding of an almost complete loss of citrate and polyamines, and a very high choline-to-creatine ratio plus an extended region of low signal intensity with capsular irregularity on T2 weighted MRI indicated a large volume of aggressive cancer. It was thereby decided that the patient should receive hormone blockade plus external beam RT without delay.

Coakley et al. [27] used 1.5T, PRESS, MRSI, TR/TE: 1000/130 ms, ratio of (choline + creatine) to citrate. 11 Scheidler et al. [16] used 1.5T, MRSI, PRESS, TR=1000 ms, TE=130 ms, frequency, phase and baseline correction, calculated integral areas of choline, creatine and citrate. 12 Yu et al. [28] used 1.5T, MRSI, PRESS, TR=1000 ms, TE=130 ms, frequency, phase and baseline correction, integral areas of choline, creatine and citrate.

10

138

For patients with negative TRUS-guided biopsy, but with rising PSA.

These patients usually have BPH, so the CG is large. This creates sampling problems using TRUS. In a preliminary study, it was found that MRI/MRSI targeting of cancer in these patients can significantly increase the positive yield of subsequent TRUS-guided biopsies (p. 458) [29]. Such a patient is presented, in whom a very small area of clear metabolic abnormality with low T2 weighted MRI signal was found, and a subsequent TRUS-guided biopsy found prostate cancer.

Guiding RT Planning A major problem for RT planning has been the differences between shape and location of the prostate with MRSI and with treatment. Wu et al. [30] describe a 3D deformable image registration method to automatically transform images from the deformed imaging state to the resting state. This should enable accurate transfer of information about regions of high tumor burden obtained with MRSI to images used for RT planning. They note: regions of absent or low citrate concentration in the prostate can be visualized at a resolution of a few mm. This new advancement provides {an} opportunity for preferential targeting of radiation to regions of high tumor burden in the prostate (p. 1577). A similar point was previously made by Tepper [31]: the ability to localize prostate cancers to defined areas of the prostate has already been demonstrated with MR spectroscopy and, with the improved dose delivery techniques already at our disposal, could markedly affect therapy. Why should the entire prostate be the target volume (at least for the boost) for a small localized lesion? (p. 547). A clinical trial is on going at the University of California San Francisco using the choline/citrate ratio to guide dose deliverance [32].
Brachytherapy Zaider et al. [34] presented a case report of a patient who had Gleason 7 clinically localized prostate cancer. The authors showed the possibility of optimizing brachytherapy on the basis of MRS in that patient. Subsequently, DiBiase et al. [33] applied MRSI13 to guide brachytherapy for localized prostate cancer in patients with favorable-risk prostate cancer (PSA 10 ng/ml, Gleason score 6, clinical stage T2a). The authors defined cancer as a citrate/(choline + creatine) ratio <1.4. They incorporated MRSI into treatment planning for 14 of 15 patients. In 1 patient MRSI revealed multifocal disease. This study illustrated the feasibility of incorporating MRSI into brachytherapy planning in patients with prostate cancer. An improved 3D algorithm has been developed using MRS positive volumes mapped to TRUS and CT for planning prostate brachytherapy [35].

Di Biase et al. [33] used 1.5T, MRSI, PRESS, TR/TE=1000/130 ms, calculated metabolite ratios on the basis of the areas of the individual peaks.

13

139

9.4

Follow-up with MRI & MRSI after treatment

MRSI evaluation of citrate is found to be highly accurate for following the progression or regression of prostate cancer after therapy [20]. Determination of the accuracy of combined MRI/3D-MRSI for detecting residual cancer after therapy is more difficult than before therapy because of the lack of a gold standard (pathology of the surgically removed prostate) and the long natural history of prostate cancer. Therefore, large-scale outcome studies are required to determine if MRI/3D-MRSI can morphologically and metabolically assess the early efficacy of chemoprevention trials (p.127) [26]. Follow-up of patients with suspected local recurrence This suspicion is most often generated by increasing PSA. However, PSA is non-specific for prostate cancer. Moreover, PSA can continue to rise for 1 to 2 years after RT (external as well as brachytherapy), and PSA is difficult to interpret in the face of hormone deprivation therapy. MRI and CT are also difficult to interpret after RT. Not even repeat biopsy is fool proof, since sampling errors can occur. MRI/MRSI can help distinguish normal or necrotic tissue from residual or recurrent prostate cancer [29].
Empirical Studies

Among 25 patients examined prior to and post-cryosurgery, histologically confirmed necrotic tissue showed a loss of all prostatic metabolites, whereas the (choline + creatine) to citrate ratios in areas of histologically confirmed BPH and prostate cancer did not change significantly from before therapy (but did differ from each other) [29]. A patient is presented with biopsy-proven prostate cancer (left lobe) with a PSA = 0.6 ng/ml at the time of MRI/MRSI, which was nearly 3 years after intensity modulated radiation therapy (IMRT). Many voxels showed complete loss of all prostate metabolites (termed metabolic atrophy) indicating effective RT. However, in the right lobe there were several voxels with (choline + creatine) to citrate ratios that were 3 Sd > normal values, and in this region biopsy confirmed the presence of residual cancer [29]. Mueller-Lisse et al. [36] used MRI and 3D MRSI14 to examine the metabolic effects of hormone-deprivation among 65 patients with prostate cancer. They found a significant time-dependent loss of choline, creatine, citrate and polyamines during hormone-deprivation therapy, leading to total metabolic atrophy (loss of all observable metabolites) in 25% of the patients receiving long-term therapy. There was a significant difference in the amount of loss and time-course of this process in normal versus cancerous prostate. Since citrate was abolished, residual prostate cancer was detected by increased
14

Mueller-Lisse, et al. [36] used 1.5T, PRESS, MRSI, TR/TE=1000/130ms, used an automatic procedure to identify significant peaks (>5:1 SNR), baseline correction, found peaks that lined up in the magnitude and real part of the spectrum and maximized the area of the peaks in the real part of the spectrum and did manual touch up of the individual spectra.

140

choline (choline to creatine ratio 1.5) or only choline in the spectrum. There was a correlation between loss of citrate and total metabolic atrophy on the one hand, and decline in PSA on the other. These authors conclude: MRI/3DMRSI provided both a measure of residual cancer and a time-course of metabolic response following hormone-deprivation therapy (p.49).

MRSI may indicate time course and mechanism of therapeutic response Consistent with the fact that prostatic citrate production and secretion is regulated by testosterone and prolactin, citrate falls dramatically after hormone blockade. There is also a time-dependent loss of all prostate metabolites with hormone therapy indicative of metabolic atrophy. Studies are on going for assessing the presence and spatial extent of prostate cancer after these therapies; the prognostic value of the time course to metabolic atrophy; the duration of undetectable metabolism, and the rate of metabolic recovery, particularly of cancer. The detection of residual cancer at an early stage following treatment, and the ability to monitor the time course of therapeutic response would allow earlier intervention with additional therapy and provide a more quantitative assessment of therapeutic efficacy (p. 460) [29].
Future perspectives

Investigators in this field have suggested a number of areas for future work with respect to MRSI and prostate cancer. Suggestions include:
Need for large scale, multi-center studies.

Single-shot echo-planar-based diffusion-weighted imaging To improve identification of prostate cancer in the peripheral zone. Apparent diffusion coefficient also may be helpful [29]. Improvement in acquisition methods. Schriker et al. [37] note that the conventional 180 pulses can have chemical shift mis-registration and require high peak power. In their paper [37], new spectral-spatial RF pulses were developed to cancel out the chemical shift misregistration including optimal phase modulation. The authors report that this is feasible and reliable for application in prostate cancer using MRSI. Addition of MRI + 3D MRSI to PSA & biopsy In order to improve patient selection and risk stratification for chemoprevention trials (i.e. to assess pre-malignancy). There is a need to find intermediate endpoints (p. 124) [26].

Wider application of 2D COSY (see next section).

141

9.5

2D COSY & in vitro MRS in prostate cancer

In vivo MRS with 2D COSY

Empirical Studies

Swanson et al. [38] examined 40 patients with untreated prostate cancer using single-voxel J-resolved spectroscopy with over-sampling in the F1 dimension. Based on MRI and biopsy information, the voxels were positioned in the peripheral zone at sites of benign or cancerous tissue, or in regions of glandular or stromal BPH within the CG. The second J-resolved dimension enabled visualization of J-modulation of citrate and resolution of the polyamines. Findings Regions of healthy tissue in the peripheral zone, and those with glandular BPH showed high citrate and polyamines, with consistent coupling and Jmodulation patterns. Cancerous tissue in the peripheral zone and stromal BPH showed low citrate and polyamines, consistent with previous in vivo and in vitro studies. Water T2 relaxation times were significantly higher in the healthy tissue compared to all of the other tissues. Conclusions This preliminary study demonstrates that J-resolved spectroscopy of the in situ prostate can be acquired, and the information obtained from the second spectral dimension can provide additional physiologic information from human prostate tissue in a reasonable amount of time (< 10 min) (p. 973). Yue et al. [39] applied localized 2D J-resolved MRS15 among 8 healthy volunteers, 3 patients with BPH and 3 patients with prostate cancer. Using this method choline containing compounds were unequivocally distinguished from spermine. The findings were consistent with those obtained ex vivo. A phantom study was also performed. These authors note that with 1D MRS or MRSI of the prostate, creatine, spermine and choline overlap. 2D MRS clearly resolved the cross peaks due to spermine. The findings are shown in Table 9.5. Assignment Citrate 2.62 2.68 ppm ( 7.9 Hz) Spermine 3.05 ppm Choline + Spermine + Creatine 3.0 3.2ppm

15

Yue et al. [39] used 1.5T, single voxel 2D J-resolved PRESS, body coil used for RF transmission, and a pelvic phased-array coil with a disposable endorectal coil for signal reception, PRESS, two-step phase cycling, TR=2000 ms, TE=30 ms.

142

Table 9.5 Metabolite Ratios obtained using in vivo 2D JPRESS MRS

(From data of Yue et al. [39])
Citrate to (Choline+Spermine+Creatine)
(Mean Sd)

Spermine to (Choline+Spermine+Creatine)
(Mean Sd)

8 Healthy Volunteers

0.71 0.2

0.42 0.16

BPH (3 patients)

0.65 0.68

0.54 0.49

Prostate Cancer (3 patients)

0.18 0.06

0.21 0.24

In vitro MRS In vitro MRS reveals other possible markers for prostate cancer; these include: polyamines, myoinositol, scylloinositol and taurine, inter alia [29].
Empirical studies

Spermine levels

Lynch et al. [40] compared 12 normal, 10 BPH, 4 prostate cancers, 11 vasal aplasias, 1 with prostatodynia. They examined prostatic fluid with in vitro MRS. Citrate and spermine were highly correlated. Relatively higher levels of spermine were found in prostate cancer for any given level of citrate, although the authors point out that the small numbers in their study preclude any definitive conclusions about this finding.
Multivariate Spectral Patterns in Prostate Cancer

Menard et al. [41] examined 116 biopsy specimens after RT. The sensitivity and specificity of MRS for identifying malignant biopsy after RT were 88.9% and 92% respectively, overall accuracy = 91.4% choline, creatine, glutamine and lipid were the diagnostic spectral regions, citrate was always absent regardless of histology. The authors concluded: although the spectral features of prostate tissue markedly change after radiotherapy, MRS combined with multivariate methods of analysis can accurately identify histologically malignant biopsies (p. 317).

143

Smith et al. [42] examined biopsies from 66 patients with BPH and 21 with prostate cancer. They used multi-variate analysis with a training set and then a test set, achieving 100% sensitivity and 95.5% specificity (overall classification accuracy 96.6%). The most discriminatory regions were centered on 3.49 (taurine), 3.43, 2.53 (citrate), 2.17, 1.87 (glutamate), and 1.15 ppm. Moreover, they were able to distinguish the two types of BPH (glandular and stromal), which require different therapy, on the basis of much higher citrate in the former. Swanson et al. [43] used HRMAS to assess 54 post-surgical prostate samples obtained using MRI/3D MRSI. Pre-surgical MRI/3D MRSI identified healthy and cancerous prostate tissues with 81% accuracy. Healthy glandular tissue was distinguished from prostate cancer by significantly higher levels of citrate and polyamines, and lower choline, phosphocholine and glycerophosphocholine. Predominantly stromal tissue lacked citrate and polyamines, but had significantly lower levels of choline compounds compared to malignant tissue. Taurine, myoinositol and scylloinositol as well as choline compounds were higher in prostate cancer. More aggressive cancers showed higher choline and lower citrate and polyamines. The authors conclude: the elucidation of spectral patterns associated with mixtures of different prostate tissue types and cancer grades, and the inclusion of new metabolic markers for prostate cancer may significantly improve the clinical interpretation of in vivo prostate MRSI data (p. 944). Swindle [44] et al. examined 77 specimens from the prostate, 61 of which were from the peripheral zone. Sensitivity was 100% and specificity 94% via in vitro MRS for identifying histologically confirmed prostate cancer. In vitro MRS distinguished BPH from adenocarcinoma with a sensitivity of 97% and specificity 88%. Depleted citrate and elevated choline levels alone were not accurate markers of malignancy, since citrate levels remain high when a small amount of malignant disease is present (p. 144). Lipid, creatine and lysine were helpful. Cancerous prostate tissue showed a significantly higher lipid to lysine ratio (1.3:1.7 ppm) compared to stromal BPH. Compared to glandular BPH, the lipid to citrate ratio (1.3:2.5 ppm) was significantly higher in malignant prostate. Spermine and spermidine (3.1 ppm) decrease with increased adenocarcinoma involvement. Although the choline-creatine ratio and lipid-lysine ratio were significantly higher in malignant prostate specimens compared to benign prostate tissue, there was considerable overlap of ratios, resulting in poor separation of BPH from cancer (p. 147). Combining these two ratios improved the diagnostic accuracy to sensitivity 100% and specificity 82%. Key conclusion: In study [44], there were prostate cancers and prostate intra-epithelial neoplasia, which were missed with routine histologic exam, but were

144

confirmed with histologic step-slice analysis. These were correctly identified using in vitro MRS.
Insights into the components of total choline

Significantly higher phosphocholine and glycerophosphocholine levels were found in human prostate cells derived from metastases compared to normal prostate epithelial and stromal cells. It is concluded: the elevation of the choline peak observed clinically in prostate cancer is attributable to an alteration of phospholipid metabolism and not simply to increased cell density, doubling time or other non-specific effects (p. 3599) [45]. Other Hypotheses

As discussed earlier, the normal prostate accumulates and secretes very high amounts of citrate, but this is lowered with malignancy, as well as post-biopsy hemorrhage, prostatitis, and therapy. There are zonal differences, with the periphery being mainly involved in producing citrate and therefore containing the highest concentrations. There has been occasional overlap between the lowest citrate levels in BPH and the highest citrate levels in prostate cancer (p. 240) [20]. Stromal tissue has low citrate levels. MRSI evaluation of citrate is found to be helpful for following the progression or regression of prostate cancer after therapy. Costello et al. [20] propose that malignant transformation of prostate cells is associated with citrate oxidation, as opposed to citrate accumulation. Zinc also accumulates at very high levels in normal prostate cells; these authors note a relationship among citrate oxidation, loss of zinc and prostate cancer.

References
[1] L.B. Signorello, H-O. Adami, Prostate Cancer, in: H-O. Adami, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, Oxford, 2002, p. 400-428. [2] P.R. Carroll, K.L. Lee, Z.Y. Fuks, P.W. Kantoff, Cancer of the prostate, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 1418-1479. [3] H.L. Scher Hyperplastic and malignant diseases of the prostate, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p.608-616. [4] E. Meuillet, S. Strattion, D. Prasad Cherukuri, A.C. Goulet, J. Kagey, B. Porterfield, M.A. Nelson, Chemoprevention of prostate cancer with selenium: an update on current clinical trials and preclinical findings, J. Cell. Biochem. 91, 443-458 (2004). [5] H.L. Parnes, M.G. House, J. Kagan, D.J. Kausal, R. Lieberman, Prostate cancer chemoprevention agent development. The National Cancer Institute, Division of Cancer Prevention portfolio, J. Urology, 171, S68-S74 (2004). [6] O.P. Heinonen, D. Albanes, J. Virtamo, et al. Prostate cancer and supplementation with alphatocopherol and beta-carotene: incidence and mortality in a controlled trial J. Natl. Cancer Inst. 90, 400-446 (1998).

145

[7] P.K. Mills, R. Yang, Prostate cancer risk in California farm workers, J. Occup. Environ. Med. 45, 249-258 (2003). [8] T.L. Weston, K.J. Aronson, J. Siemiatycki, G.R. Howe, L. Nadon, Cancer mortality among males in relation to exposures assessed through a job-exposure matrix, Int. J. Occup. Environ. Health 6, 194-202 (2000). [9] I.M. Thompson, D.K. Pauler, P.J. Goodman et al. Prevalence of prostate cancer among men with a prostate-specific antigen level 4.0 ng per milliliter, N Engl. J. Med. 350, 2239-2246 (2004). [10] E. Salminen, A. Hogg, D. Binns, et al. Investigations with FDG-PET scanning in prostate cancer show limited value for clinical practice, Acta Oncol. 41, 425-429 (2002). [11] M. al-Abany. Towards elimination of anal-sphincter and rectal dysfunction after radiation therapy for prostate cancer, [Doctoral Dissertation], Stockholm, Karolinska Institute, 2004. [12] D.K. Ornstein, J. Kang, How to improve prostate biopsy detection of prostate cancer, Curr.Urology Rep. 2, 218-223 (2001). [13] A.E. Li, D.A. Bluemke, Magnetic resonance imaging in V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 669-679. [14] Y. Kaji, A. Wada, I. Imaoka, et al. Proton two-dimensional chemical shift imaging for evaluation of prostate cancer: external surface coil vs. endorectal surface coil, J. Magn. Reson. Imaging, 16, 697-706 (2002). [15] M.D. Macilquham, J. Gong, A.M. Lavoipierre, MR widens options for prostate imaging, Diag. Imaging Eur. 37-45 (2003). [16] J. Scheidler, H. Hricak, D.B. Vigneron, et al. Prostate cancer: localization with threedimensional proton MR spectroscopic imagingClinicopathologic study, Radiology 213, 473-480 (1999). [17] G.P. Liney, L.W. Turnbull, M. Lowry, L.S. Turnbull, A.J. Knowles, A. Horsman, In vivo quantification of citrate concentration and water T2 relaxation time of the pathologic prostate gland using 1H MRS and MRI, Magn. Reson. Imaging 15, 1177-1186 (1997). [18] F.V. Coakley, A. Qayyum, J. Kurhanewicz, Magnetic resonance imaging and spectroscopic imaging of prostate cancer, J. Urology, 170, S69-S76 (2003). [19] X. Wu, S.J. Dibiase, R. Gullapalli, C.X. Yu, Deformable image registration for the use of magnetic resonance spectroscopy in prostate treatment planning, Int. J. Radiation Oncology Biol. Phys. 58, 1577-1583 (2004). [20] L.C. Costello, R.B. Franklin, P. Narayan, Citrate in the diagnosis of prostate cancer, Prostate, 38, 237-245 (1999). [21] R. Dhingsa, A. Qayyum, F.V. Coakley, Y. Lu, K.D. Jones, M.G. Swanson, P.R. Carroll, H. Hricak, J. Kurhanewicz, Prostate cancer localization with endorectal MR imaging and MR spectroscopic imaging: effect of clinical data on reader accuracy, Radiology 230, 215-220 (2004). [22] J.S.P. Yuen, C.H. Thng, P.H. Tan, et al., Endorectal magnetic resonance imaging and spectroscopy for the detection of tumor foci in men with prior negative transrectal ultrasound prostate biopsy, J. Urology 171, 1482-1486 (2004). [23] K.L. Zakian, S. Eberhardt, H. Hricak, et al. Transition zone prostate cancer: metabolic characteristics at 1H MR spectroscopic imaginginitial results, Radiology 229, 241-247 (2003). [24] J.M. Garcia-Segura, M. Sanchez-Chapado, C. Ibarburen, J. Viano, J.C. Angulo, J. Gonzalez, J.M. Rodriguez-Vallejo, In vivo proton magnetic resonance spectroscopy of disease prostate:

146

spectroscopic features of malignant versus benign pathology, Magn. Reson. Imaging 17, 755-765 (1999). [25] G.P. Liney, L.W. Turnbull, A.J. Knowles, In vivo magnetic resonances spectroscopy and dynamic contrast enhanced imaging of the prostate gland, NMR Biomed. 12, 39-44 (1999). [26] J. Kurhanewicz, M.G. Swanson, P.J. Wood, D.B. Vigneron, Magnetic resonance imaging and spectroscopic imaging: improved patient selection and potential for metabolic intermediate endpoints in prostate cancer chemoprevention trials, Urology, 57 (Suppl 4A), 124-128 (2001). [27] F.V. Coakley, J. Kurhanewicz, Y. Lu, K.D. Jones, M.G. Swanson, S.D. Chang, P.R. Carroll, H. Hricak, Prostate cancer tumor volume: measurement with endorectal MR and MR spectroscopic imaging, Radiology, 223, 91-97 (2002). [28] K.K. Yu, J. Scheidler, H. Hricak, et al., Prostate cancer: prediction of extracapsular extension with endorectal MR imaging and three-dimensional proton MR spectroscopic imaging, Radiology 213, 481-488 (1999). [29] J. Kuranewicz, M.G. Swanson, S.J. Nelson, D.B. Vigneron, Combined magnetic resonance imaging and spectroscopic imaging approach to molecular imaging of prostate cancer J. Magn. Reson. Imaging 16, 451-463 (2002). [30] X. Wu, S.J. Dibiase, R. Gullapalli, C.X. Yu, Deformable image registration for the use of magnetic resonance spectroscopy in prostate treatment planning, Int. J. Radiation Oncology Biol. Phys. 58, 1577-1583 (2004). [31] J. Tepper, Form and function: The integration of physics and biology. Int. J. Radiation Oncology Biol. Phys. 47, 547-548 (2000). [32] C.C. Ling, J. Humm, S. Larson, et al., Towards multidimensional radiotherapy (MD-CRT): biological imaging and biological conformality, Int. J. Radiation Oncology Biol. Phys. 47, 551560 (2000). [33] S.J. DiBiase, K. Hosseinzadeh. R. P. Gullapalli, et al., Magnetic resonance spectroscopic imaging-guided brachytherapy for localized prostate cancer, Int. J. Radiation Oncology Biol. Phys. 52, 429-438 (2002). [34] M. Zaider, M.J. Zelefsky, E.K. Lee et al. Treatment planning for prostate implants using magnetic-resonance spectroscopy imaging, Int. J. Radiation Oncology Biol. Phys. 47, 1085-1096 (2000). [35] T. Mizowaki, G.N. Cohen, A.Y. Fung, M. Zaider, Towards integrating functional imaging in the treatment of prostate cancer with radiation: the registration of the MR spectroscopy imaging to ultrasound/CT images and its implementation in treatment planning, Int. J. Radiation Oncology Biol. Phys. 54, 1558-1564 (2002). [36] U.G. Mueller-Lisse, M.G. Swanson, D.B. Vigneron, et al., Time-dependent effects of hormone-deprivation therapy on prostate metabolism as detected by combined magnetic resonance imaging and 3D magnetic resonance spectroscopic imaging, Magn. Reson. Med. 46, 49-57 (2001). [37] A. A. Schriker, J.M. Pauly, J. Kurhanewicz, M.G. Swanson, D.B. Vigneron, Dualband spectral-spatial RF pulses for prostate MR spectroscopic imaging, Magn. Reson. Med. 46, 10791087 (2001). [38] M.G. Swanson, D.B. Vigneron, T-K C. Tran, N. Sailsuta, R.E. Hurd, J. Kurhanewicz, Singlevoxel oversampled J-resolved spectroscopy of in vivo human prostate tissue, Magn. Reson. Med. 45, 973-980 (2001). [39] K. Yue, A. Marumoto, N. Binesh, M.A. Thomas, 2D JPRESS of human prostates using an endorectal receiver coil, Magn. Reson. Med. 47, 1059-1064 (2002). [40] M.L. Lynch, J.K. Nicholson, Proton MRS of human prostatic fluid: correlations between citrate, spermine and myoinositol levels and changes with disease, Prostate 30, 248-255 (1997).

147

[41] C. Menard, I.C.P Smith, R.L. Somorjai, et al., Magnetic resonance spectroscopy of the malignant prostate gland after radiotherapy: a histopathologic study of diagnostic validity, Int. J. Radiation Oncology Biol. Phys. 50, 317-323 (2001). [42] I.C. Smith, D.E. Blandford, Diagnosis of cancer in humans by 1H NMR of tissue biopsies, Biochem. Cell Bilo. 76, 472-476 [43] M.G. Swanson, D.B. Vigneron, Z.L. Tabatabai, et al. Proton HR-MAS spectroscopy and quantitative pathologic analysis of MRI/3D-MRSI-targeted postsurgical prostate tissues, Magn. Reson. Med. 50, 944-954 (2003). [44] P. Swindle, S. McCredie, P. Russell, Pathologic characterization of human prostate tissue with proton MR spectroscopy, Radiology 228, 144-151 (2003). [45] E. Ackerstaff, B.R. Pflug, J.B. Nelson, Z.M. Bhujwalla, Detection of increased choline compounds with proton nuclear magnetic resonance spectroscopy subsequent to malignant transformation of human prostatic epithelial cells, Cancer Res. 61, 3599-3603 (2001).

149

Chapter 10

Gynecologic Cancers
_______________________________________________________________________________

In this chapter we will review the three main gynecologic malignancies: ovarian cancer, cancer of the uterine cervix and endometrial cancer.

10.1 Ovarian Cancer

10.1.1 Overview of epidemiological and clinical aspects
There are three major clinical pathological entities: epithelial carcinomas, germ cell tumors and stromal carcinomas. Approximately 90% of ovarian cancers in the U.S. are of the epithelial type [1]. Incidence and prevalence/morbidity and mortality An estimated 191 000 new cases of ovarian cancer are diagnosed each year worldwide [2]. It is the most common cause of death from gynecologic cancer in the U.S., accounting for about 5% of all cancer deaths among women [3]. In 1999, approximately 25 000 new cases were diagnosed and 14 500 women in the U.S. died of ovarian cancer [1]. High incidence rates are also reported in Northern Europe, in particular in Sweden and Poland, as well as in the U.K. [4]. Rates of ovarian cancer are rising in countries with previously low reported incidence, such as Japan. Higher rates are reported among younger Chinese-American and Japanese-American women compared to native-born Chinese and Japanese women [4].

150

Epithelial tumors are rare before the age of 40, and there is an increasing incidence with age, peaking in the eighth decade [3].

Etiology/risk factors Inherited susceptibility Familial cases account for approximately 5% of all ovarian cancer. A positive family history is considered a major risk factor. While in the general population the lifetime risk of developing ovarian cancer is 1.6%, this rises to 5% if there is an affected 1st degree relative. If 2 or more affected 1st degree relatives, the risk may be more than 50% [3, 4]. The three major inherited syndromes are:
. Ovarian cancer only

Ovarian and breast cancer (BRCA 1 and BRCA 2) Lifetime risk is estimated to between 16 and 65% [5]. This syndrome accounts for 85-90% of all hereditary ovarian cancers that are identified. Most are BRCA 1 [1]. Non-polyposis colorectal cancer, endometrial cancer and ovarian cancer

Gynecologically related risk factors

Nulliparity Well-established risk factor. The hypothesis of incessant ovulation implies a defective repair process of the surface epithelium as the key to development of ovarian cancer. Pregnancy reduces the risk of ovarian cancer [2, 3]. The use of fertility drugs has not been clearly demonstrated to be associated with increased risk; however, they may adversely affect some subgroups with respect to ovarian cancer risk [6]. Unopposed estrogen hormone replacement therapy From meta-analysis, the OR = 2.56 (95% CI = 1.32 4.94) for the use of hormone replacement therapy (HRT) with unopposed estrogen [2]. Polycystic ovary syndrome Associated with increased risk of ovarian, as well as endometrial carcinoma [2, 7].

Gynecologically related protective factors

Large number of full-term pregnancies The OR is 0.17 (95% CI = 0.05 0.54) for 7 full-term pregnancies compared to nulliparous women with respect to ovarian cancer [2].

151

Breast-feeding Oral contraceptives

Protection conferred by these factors is considered to be due to suppression of ovulation.

Other lifestyle related factors

Obesity / High fat diet Correlation studies suggest a possible role for increased fat and diary products, e.g. increased westernisation of the Japanese diet. The role of dietary fat is also implicated by migration studies, indicating increased ovarian cancer risk for Japanese women after moving to the U.S. A speculated mechanism is that high animal fat diets could lead to estrogen production by gut bacteria [4]. Obesity is reportedly associated with an OR = 1.7 (95% CI = 1.1 2.8) [2], although there are some contradictory findings [4].

Other Medical Conditions and Related Exposures

Radiation therapy for cervical cancer This confers after 10 years an estimated relative risk = 1.4 [4].

Occupational and Environmental Factors Epidemiological studies examining the association between occupational and environmental factors and ovarian cancer have had numerous methodological difficulties [8]. The need for well-designed analytic epidemiological studies with sufficient power is underscored by Shen et al. (p. 175) [9].
Ionizing radiation A reported two-fold increased risk of ovarian cancer among women in Hiroshima and Nagasaki exposed to >100 rads compared to non-exposed women. Risk particularly for women who were in the reproductive years in 1945 [4]. Occupations and work-related exposures In a study of Mortality Odds Ratios among female health care workers in 24 U.S. states from 1984-1993 from death certificates, a 30% excess ovarian cancer was found in nurses (30%). The authors note the need for further studies with data on reproductive history [10]. There is some evidence of an association between general exposure to pesticides and to those containing triazine and risk of ovarian cancer [11, 12].

152

A study [8] using data linkage for occupationally active Finnish women found increased risks of ovarian cancer associated with exposure to aromatic hydrocarbon solvents (standardized incidence ratio (SIR) = 1.3 (95% CI = 1.0 1.7), and gasoline (SIR = 1.5 (95% CI = 1.0 2.0)). The data indicated an increased risk for hairdressers and women in the printing industry, but not dry cleaning. A review of registry data on Swedish women indicates increased risk among women in dry cleaning, telegraph and telephone work, paper packaging, graphic and printing work [13]. No increased risk was found among hairdressers and beauticians. Organic dusts, aromatic amines, aliphatic and aromatic hydrocarbons are suggested as specific etiologic agents (p. 200). The authors note that this study confirms some earlier findings on smaller cohorts and identified some new relationships [13]. Other studies on pulp, paper and printing industry In the Norwegian pulp and paper industry compared to the general female population there was an excess risk of ovarian cancer. The specific incidence rate = 1.5 (95% CI = 1.07 2.09). The authors suggest a possible link to talc, microbes and/or paper dust [14]. Among women in the Russian printing industry there was an increased risk among bookbinders (standard mortality ratio = 2.9, 95% CI = 1.5 - 5.0). There was also one death from mesothelioma of the abdomen. The authors consider this may be related to use of asbestos-contaminated talc fillers in paper [15]. A study of women undergoing incidental oophorectomy demonstrates that even with second-hand exposure (household contact with an asbestos worker), asbestos can reach the ovary [16].

Clinical Presentation If localized, ovarian carcinoma is usually asymptomatic. Rarely, it may be associated with vaginal bleeding. Symptoms such as abdominal pain, bloating and urinary symptoms usually indicate advanced stage. Ovarian cancer may present as a palpable adnexal mass during routine pelvic examination. However, if the woman is still menstruating this is much more often a benign functional cyst [3] (see next sub-section). Differential Diagnosis of an adnexal mass There is a wide diversity of both benign and malignant ovarian tumors. Some of the more common include:

Physiological enlargement Common during the reproductive years. Benign functional cyst Common during the reproductive years.

153

Endometriomas

Metastic disease to the ovary Difficult to differentiate from a primary ovarian malignancy, yet surgical treatment and chemotherapy may vary greatly (p. 604) [5].
Pedunculated leiomyomas Germ cell tumors (95% benign)

Inflammatory masses These can include chronic tubo-ovarian abscesses, actinomycosis, and tuberculosis. Classification, Grading and Associated Prognosis Ovarian epithelial cancers may be either of low malignant potential showing the histopathologic features of malignancy, but not invading the ovarian stroma, or they may be invasive. The former are more common among younger women. There are five different types of malignant epithelial ovarian tumors: serous (50%), mucinous (25%), endometroid (15%), clear cell (5%) and Brenner tumors (1%). Approximately 4% of ovarian tumors are stromal or germ cell type; these are managed similarly to testicular cancer in men. Histologic grade is important with respect to prognosis of epithelial ovarian tumors [3, 4]. The finding of increased tumor p53 is a poor prognostic sign [1, 2]. Staging As described in [1], staging of ovarian cancer is as follows:
Stage I = Confined to ovary, 90% 5-year survival Ia: Limited to 1 ovary, no tumor on external surface, no ascites, capsule intact Ib: Both ovaries, no tumor on external surface, no ascites, capsule intact Ic: Ia or Ib, but tumor on surface or capsule rupture, or ascites, or positive peritoneal washings Stage II = Confined to pelvis, 70% 5-year survival IIa: Extension to uterus or tubes IIb: Pelvic extension IIc: IIa or IIb, but tumor on surface or capsule rupture, or ascites, or positive peritoneal washings Stage III = Intra-abdominal spread, 15-20% 5-year survival Stage IV = Outside abdomen, 1-5% 5-year survival

154

According to Coakley [17], up to 90% of patients with apparent stage I or II ovarian cancer do not have optimal surgical staging, often because of failure to perform a selective retroperitoneal lymphadenectomy. As a result, approximately 30% of such patients are under-staged (p.630).

Treatment
Barakat and Hricak [18] state; one of the most important prognostic factors in epithelial ovarian cancer is the volume of disease that remains after surgical cytoreduction (p. 524). For stage I, well or moderately differentiated and no residual after surgery, no further therapy is given. For this group the 5-year survival is over 95%. For all other patients, adjuvant therapy probably is usually recommended (cisplatin or platinum-containing combinations). For stromal tumors, surgery only is the usual treatment. For Germ cellsurgery plus chemotherapy is recommended [3]. Ovarian sarcomas (rare tumors) show improved outcome with cytoreduction to a residual tumor burden of 1 cm. [19]. Since ovarian cancer is sensitive to chemotherapy, patients with small-volume residual disease can often be put into either long-term remission or even cure. There is a very wide variation in the percentage of patients who undergo optimal cytoreduction. If not, then the primary surgical procedure does not substantially affect overall survival. There is still no preoperative test to accurately predict optimal versus sub-optimal cytoreduction in patients with advanced ovarian cancer. It has been noted that imaging pre-operatively can help predict which patients may not be amenable to optimal debulking and instead would likely benefit more from neo-adjuvant chemotherapy. Postoperatively, imaging can provide objective confirmation of the surgical assessment of residual disease, and thereby more accurately perform stratification of patients for clinical trials [18].

10.1.2 Current approach to 1 diagnosis & staging

Ovarian cancer is usually diagnosed when it has already spread beyond the true pelvis [3]. Thus, epithelial cancers of the ovary have been described as a silent killer because the overwhelming majority of patients present with disease that has spread outside of the ovary and indeed outside of the pelvis at the time of initial presentation (p. 1598) [1]. Pelvic examination

155

Pelvic exam has severe limitations for identifying adnexal pathology. Sensitivity for detection of left adnexal masses has ranged between 2336% and for right adnexal masses 15-28%. Attending physicians and residents have been found to perform only slightly better than medical students [20]. CA-125 Overall, CA-125 is positive in 80-85% of patients with epithelial ovarian cancer [3]. However, it has poor sensitivity for detection of early stage ovarian cancer, being expressed in only 50% of Stage I cancers [5]. CA-125 can also be elevated with other malignancies: endometrial carcinoma, as well as cervical, fallopian, pancreatic, breast, lung and colon cancers. Moreover, a positive CA-125 is seen with pregnancy, endometriosis, pelvic inflammatory disease and uterine fibroids [3]. Some key information for imaging in ovarian cancer diagnostics
The germinal epithelium is a single layer of columnar cells that line the ovary (recall that about 90% of ovarian cancers are epithelial in origin). Epithelial cancers are usually cystic, with irregular internal solid components. Because of the mobility of the ovaryovarian masses can be seen not only lateral to the uterus, but also midline. Ovarian cancer is often bilateral (multicentric origin) or less commonly from metastatic spread from the other ovary. Calcification suggests serous tumors, but only 12% of serous tumors have visible calcification (as seen on CT) [17].

Transvaginal sonography (TVS) A change in ovarian size or volume may be an early indication of ovarian cancer. The upper limit of normal for pre-menopausal ovaries is 20 cm3 and 8-10 cm3 for ovaries after menopause. With age, the ovaries normally become even smaller. Any ovary enlarged for age or exceeding twice the volume on the contralateral side is considered suspicious by sonographic criteria (p. 592) [5]. Multiple logistic regression reveals that the two most significant factors in distinguishing benign from malignant adnexal

156

mass on TVS are the presence of solid elements in the tumor, and presence of central vascular flow [20]. Most cancerous ovarian masses have central vessels in irregular areas of mural thickening and within papillary projections, whereas benign masses usually have peripheral vessels with a regular branching pattern. Transvaginal sonography is being evaluated for population screening. Close surveillance and screening with TVS is used for women at high risk. However, there is no evidence that this lowers mortality [1]. A major shortcoming of TVS is that it yields mainly false positive results. Young [3] cites a study in which 67 laparotomies were required to diagnose one primary ovarian cancer.
Doppler flow imaging Doppler flow imaging can improve the diagnostic accuracy of TVS. Adding color Doppler increased positive predictive value from 63% to 97%. Combined techniques provide the best sensitivity and specificity (estimate 92%) [5]. Pulsed Doppler: Low resistive index (RI)1 <0.4 considered abnormal. Pulsatility index (PI)2 < 1 considered abnormal. While Pulsed Doppler cannot be used as an independent indicator of malignancy, it has been found to offer supplementary information to help distinguish benign from malignant ovarian masses [5]. As noted by Cho et al. [21]: Most adnexal lesions manifest as cystic masses and differential diagnosis on the basis of imaging is difficult. Ultrasonography with color Doppler is currently the standard imaging tool for the differential diagnosis of adnexal masses, and although very sensitive in detecting these and sometimes able to indicate malignancy, it is frequently unable to provide images adequate for specific diagnosis (p. 105).

Multi-modal screening Anderiesz and Quinn [22] recommend screening women at high risk annually with CA-125 + TVS. Multimodal screening with serum CA125 and transvaginal ultrasonography has been shown to improve survival. However, the results so far do not justify routine screening until the impact of screening on mortality has been assessed in larger randomised trials (p. 210) [23]. This is a widely held view. The importance of risk stratification in multi-modal screening is underscored by Paley [24], who states: To date, no evidence exists
1 2

RI = {(peak systolic velocity) (end-diastolic velocity)}/ (peak systolic velocity) PI = {(peak systolic velocity) (end-diastolic velocity)}/(mean velocity)

157

to justify mass population screening; however, emphasis on subpopulation screening, identification of novel serologic markers, and multi-modality screening designs have provided hope that these refinements may eventually translate into a reduction in ovarian cancer mortality (p. 399). MRI The value of MRI for the evaluation of adnexal masses is increasingly appreciated. Togashi [25] states: The non-invasive nature of MR imaging is beneficial in evaluations of what are probably benign disease in young women of reproductive age In women of reproductive age, normal ovaries were identified in 82 of 84 cases on MR imaging (p. 799). During pregnancy MRI is thought to be safe, but a general recommendation is to wait until week 12 (see Chapter 6). MRI is considered superior to CT for diagnosis of malignant ovarian masses and has been shown to increase the specificity for diagnosis of malignancy in adnexal masses considered suspicious by TVS. However, the main challenge to the radiologist is to differentiate benign from malignant adnexal masses. Both ultrasound and MRI perform well for prediction of benignity. There is less specificity for diagnosis of malignancy, but features such as papillary projections, thickened septations, and internal vascularity within nodules aid in this differentiation For cases considered suspicious by TVS, more specific diagnosis by MRI may obviate the need for surgery or otherwise change management by identification of benign etiology (p. 605)[5].
Technical Considerations There is varying uniformity among pelvic MRI protocols because of the lack of evidence-based criteria (p. 599) [5]. A pelvic phased array coil or body coil is most often used. Contrast-enhancement with gadolinium improves characterization of lesions. T2 weighted imaging shows the zonal anatomy of the ovary: the lower intensity cortex and higher intensity medulla. Benign lesions on MRI: MRI can often provide the specific diagnosis of benign ovarian masses (pedunculated leiomyomas, dermoid cysts and endometriomas, with accuracy > 90%) [5]. Homogeneous, low signal intensity on T1 weighted images and high signal intensity on T2 weighted images are usually benign cysts. Cog-wheeling on T2 weighted images, or serpiginous configuration suggests hydrosalpinx. High signal intensity on T1 weighted images usually contains blood or fat.

158

Leiomyomas are usually low signal intensity on T1 and T2 weighted images. Stalk visualization is pathognomonic of a pedunculated leiomyoma. Heterogeneous benign masses can mimic malignancy Malignant-appearing lesions on MRI: Diagnostic accuracy of CE-MRI ranges from 87-99%, when a solid, enhancing lesion is found. Criteria for malignancy: Size > 4 cm Solid mass or large solid component Wall thickening > 3 mm Septa > 3 mm Vegetation, nodularity and necrosis. Necrosis is considered to be an important indicator of malignancy in a solid lesion. This is better visualized with contrast-MRI than CT. CE can reveal solid elements not seen on T2 weighted or pre-contrast T1 images. MRI cannot histologically differentiate specific surface epithelial, germ cell, stromal cell or metastatic tumors Metastic disease to the ovary is difficult to differentiate from a primary ovarian malignancy, yet surgical treatment and chemotherapy may vary greatly (p. 604) [5]. An initial study [26] suggests that diffusion imaging may further help to differentiate benign and malignant ovarian masses.

10.1.3 In vivo proton MRS in ovarian cancer diagnostics

Thus far, there have been very few investigations applying proton MRS in vivo to evaluate ovarian masses. Currently, it is still technically difficult to obtain good quality spectra due to motion artifacts from respiratory and peristaltic movements [21]. Cho et al. [21] applied proton MRS3 to examine 31 patients with adnexal lesions that were subsequently treated surgically. A total of 7 patients had malignant ovarian tumors (3 were metastatic). The remaining 24 patients had benign lesions of various types (cystadenoma, teratoma, endometriosis, ectopic pregnancy, salpingitis with necrosis). The lesions were between 4 and 23 cm (in all cases larger than the localization of the voxel). The resonance at 1.3 ppm attributed to the methylene group in fatty acid chains was observed in 5 of 7 patients with malignant ovarian tumors, in 7 of 11 benign teratomas, and in the case of salpingitis. This resonance at 1.3 ppm was not seen in any of the 6 benign ovarian epithelial tumors. The lipid
3

Cho et al. [21] used 1.5T, localization of voxel via MRI or CT. Voxel size = 8cm3. STEAM, TR/TE: 3000/30ms. No respiratory gating. Post-processing via Lorentzian-to-Gaussian transformation with zero filling.

159

peak at 5.2 ppm attributed to the olefine group in fatty acid chains was seen in 2 of the 11 cases of benign teratomas, 1 of the 4 cases of endometriosis, and the ectopic pregnancy, but was not seen in any of the malignant ovarian tumors nor in any of the benign epithelial tumors. The findings from Ref. [21] are summarized in Table 10.1. Table 10.1 Lipid Peaks on In Vivo MRS of Ovarian Cancers, Non-Malignant Processes Affecting the Ovary (From data of Cho et al. [21])
Lipid Peak
1.3 ppm Malignant Ovarian Tumor Benign Epithelial Tumor Borderline malignant mucinous cystic tumor Endometriosis 5 of 7 patients

Lipid Peak
5.2 ppm 0 of 7 patients

Lipid/Water Ratio
Mean sd (Range) 0.007 001 (0.006 0.009) 0.007 0.002 (0.005 0.007) 0.006 (0.006) 0.006 0.001 (0.004 0.007) 0.101 0.138 (0.004 0.341) 0.010

0 of 6 patients

0 of 6 patients

Present in the 1 patient 0 of 4 patients

Absent in the 1 patient 1 of 4 patients

Benign Teratoma

7 of 11 patients

2 of 11 patients

Salpingitis

Present in the 1 patient Absent in the 1 patient

Absent in the 1 patient Present in the 1 patient

Ectopic Pregnancy

0.010

p =0.016, Fishers exact test (1 sided), p < 0.05 (2-sided) Lipid/water ratio was assessed in only 4 of the 7 patients with malignant ovarian tumors, 4 of the 6 patients with benign epithelial tumors.

Proton MRS4 was also used by Okada et al. [27] in 23 patients with adnexal masses. Subsequent histopathologic examination revealed primary ovarian carcinoma in 5 patients, 1 patient with endometrial carcinoma to the ovary, 10 patients with benign primary ovarian tumors, 5 patients with benign tumor of uterine origin and 2 patients with benign tumor of other origin (Schwannoma, paraganglioma). The tumor diameters ranged from 5 20 cm (mean 10.6 4.7 cm).
4 Okada et al. [27] used 1.5T, single voxel (within the tumor to avoid contamination from surrounding normal tissue), voxel size = 18-27 mL, PRESS, TE=135 ms, (except in one patient with a chocolate cyst evaluated at TE=18ms with STEAM to assess a substance with a short TE, and then with PRESS TE=135 ms). Metabolite concentrations assessed semi-quantitatively in comparison to noise (+), (++), and (-), the latter meaning the same as the average noise level. Lactate seen as an inverted doublet.

160

Their results are summarized in Table 10.2. The authors note the need to avoid contamination of peripheral tissue and fat. The high lipid peak in dermoid cysts is consistent with the fact that these are known to have a high fat content. They could not identify the 2.2 to 2.5 ppm peak, but considered that it is most consistent with glutamate or glutamine. The authors conclude: clinical proton MRS may provide useful information for the diagnosis of female intrapelvic tumors, and further clinical evaluation and in vitro confirmation are required to establish the usefulness of proton MRS in female pelvic tumors (p. 916). Table 10.2
Semi-Quantitative Assessment of Metabolites Using In Vivo MRS in Malignant vs Benign Tumors of the Ovary (From data of Okada et al. [27])
Choline
Ovarian Cancer (++) 2 patients (--) 3 patients

Creatine
(--) 5 patients

Lactate
(++) 3 patients (+) 2 patients*

Lipid
(--) 4 patients (+) 1 patient*

Other
(+) 2.1 ppm in 2 patients

Uterine Origin

(++) 1 patient

(+) 1 patients

(++) 1 patient

(--) 1 patient

Benign Ovarian Origin

(++) 2 patients (--) 3 patients

(--) 4 patients

(--) 3 patients (- +) 1 patient

(--) 3 patients (++) 1 patient

Benign Other Origin

(++) 4 patients

(+ ++) 2 patients (--) 2 patients

(--) 3 patients (- +) 1 patient

(--) 3 patients (++) 1 patient

(+) 2.2-2.5 ppm in 2 patients

*In 1 patient lactate was mixed with lipids

10.1.4 In vitro MRS in ovarian cancer diagnostics

Substantially more investigation of malignant and benign ovarian lesions using MRS has been performed in vitro. Using peak amplitude ratios, Smith and Blandford [28] were able to distinguish normal and benign from borderline and malignant ovarian samples with 95% sensitivity, and 86% specificity. They employed linear discriminant analysis training using leave-one-out (12 normal, 22 cancer) for analysis of 7 normal and 15 cancer specimens.

161

The discriminating resonances were:

1.47 ppm (fatty acid) 1.68 ppm (lysine) 2.80 ppm (fatty acid) 2.97 ppm (creatine) 3.17 ppm (choline) 3.34 ppm (taurine).

Wallace et al. [29] evaluated 19 normal, 3 borderline, and 37 ovarian carcinomas, based upon peak amplitude ratios of the following resonances:
0.9 ppm = lipid methyl 1.3 ppm = lipid methylene 1.7 ppm = lysine and polyamines 3.2 ppm = choline.

Normal/benign samples could be distinguished from borderline and malignant samples with a sensitivity of 95% and specificity of 86%. A very detailed comparison of the spectroscopic features of ovarian cancer versus non-malignant cysts is provided by Boss et al. [30]. Ovarian cyst fluid samples were taken from 40 patients (12 with malignant and 28 with benign tumors5). 1D and 2D COSY analyses were performed. Table 10.3 summarizes the significant differences between malignant and benign cysts. While there are numerous significant differences in metabolic concentrations between the cancerous and benign cysts, the ranges were very wide and overlapping. Moreover, many resonances were unassigned; for example, singlets at 2.25 and 2.77 ppm that seemed to be specific for malignant ovarian cysts.

Five of the benign lesions were not included in the analysis (3 endometriomas and 2 benign cystic teratomas).

162

Table 10.3 Significant Differences in in vitro Spectroscopic Features of Malignant and Benign Ovarian Cysts (From data of Boss et al. [30])
Malignant Ovarian Cysts (12 patients) in mol/L Metabolite (ppm) Isoleucine (1.02)
Median (Range) Significance

Benign Ovarian Cysts in mol/L (23 patients)

Median (Range)

79 (45 1398) 395 (225 644) 248 (169 541) 586 (32 -- 1352) 6536 (3426--22,214) 490 (372966) 62 (< 5129) 828 (5071352) 42 (19 113)

10 (< 10 136) 113 (< 10 586) 90 (< 10 406) 293 (< 10 924) 2479 (631-11,944) 101 (< 15 1149) 7 (< 5 90) 275 (< 30 811) 15 (< 20 163)

Valine (1.04)

Threonine (1.33)

Alanine (1.51)

Lactate (1.41)

Lysine (1.67 1.78)

Methionine (2.13)

Glutamine (2.42 2.52) Choline (3.19)

Mann-Whitney test: p < 0.001, p < 0.01, p < 0.05.

Results from a study by Massouger et al. [31] of fluid samples from 9 malignant and 19 ovarian cysts also showed higher lactate, isoleucine, valine, methionine and alanine in the cancerous specimens, but again generally with wide, overlapping ranges. In addition, these authors found higher 3-hydroxybutyrate and pyruvic acid in the malignant cyst fluid. They note that rapid cellular metabolism will lead to elevated 3hydroxybutyric acid. The high concentrations of branched chain amino acids (isoleucine, leucine, valine) are seen as protein breakdown products due to necrosis and proteolysis. Also included in their study was an endometrioma, the fluid from which showed much higher levels of high isoleucine, valine, threonine,

163

alanine, lysine, methionine, and glycine than did the malignant cysts. A mature teratoma had choline levels 350 times higher than in benign cyst fluid. It has been proposed that aspiration of ovarian cystic fluid might be performed preoperatively for diagnostic purposes. However, this entails a risk of of seeding malignant cells from an intact early ovarian carcinoma. Massouger et al. [31] contend that the levels of a number of metabolites in fluid from malignant cysts appear to be high enough to be detected with in vivo proton MRS, and therefore consider this to be the method of choice, insofar as the current obstacles hindering the acquisition of high quality spectra can be surmounted.

10.2 Cancer of the Uterine Cervix

10.2.1 Overview of epidemiological and clinical aspects
Carcinoma of the cervix was previously the most common cause of cancer death among women. Over the past 30 years due to widespread screening using the Papanicolaou6 smear, deaths rates have fallen by 50%. Incidence and prevalence/morbidity and mortality Cancer of the uterine cervix is the third most frequently occurring malignancy among women worldwide. Nearly 80% of cases occur in the developing countries, where this is one of the most common cancers among women. Overall 5-year survival is 50%, but is higher in low risk countries, due mainly to earlier detection and adequate treatment. The highest incidence is during middle age. Since this is a cancer diagnosed among middle-aged and younger women, the total years of life lost due to this malignancy are substantial (p. 342) [32]. Etiology/risk factors Infection
Human Papilloma Virus (HPV)

HPV infection is sexually transmitted and associated with anogenital warts (condylomata acuminata).
6

Often called Pap smear for short

164

HPV is considered the most important risk factor for cancer of the uterine cervix [32]. HPV infection of the lower genital tract can be associated with condylomatous atypia of the cervix, which can progress to cervical intraepithelial neoplasia (CIN), which precedes invasive cervical carcinoma. Specific subtypes associated with cervical cancer are: 16, 18, 31, 45, 51-53. HPV attacks the G1 checkpoint of the cell cycle, its E7 protein binds and inactivates Rb protein, and E6 evokes p53 degradation [3].

Gynecologically related risk factors

Multiparity Considered to be independent of HPV and other reproductive and sexual behavior variables, although multiple pregnancies might increase HPV persistence and oncogenic action by allowing the virus to evade immunological control [32].

Exposures related to lifestyle

Smoking After adjustment for HPV infection, smoking is associated with CIN, carcinoma in situ and invasive cervical cancer. Polyaromatic hydrocarbons from cigarette smoke can inhibit cell proliferation of normal cervical cell lines in vitro. Both nicotine and cotinine have been found in cervical secretions from female smokers [32].

Occupational Factors A multi-center case-control study that adjusted for sexual behavior, Pap smear screening behavior, and smoking found increased risk in some groups of service and industrial workers (maids, cleaners and cooks) [33]. Clinical Presentation Carcinoma in situ may be totally asymptomatic. Typical symptoms include abnormal bleeding or post-coital spotting, yellow vaginal discharge, lumbosacral back pain and urinary symptoms [3].

Histopathology, Stages, Treatment and Prognosis The vast majority (about 80%) of invasive cancers of the uterine cervix are squamous cell carcinomas, 10-15% are adenocarcinomas, 2-5% are adenosquamous and 1-2% are clear cell mesonephric tumors [3].

165

Stage 0 = Carcinoma in situ Treated with cone biopsy or hysterectomy, 100% 5-year survival, Stage I = Confined to uterus Treated with radical hysterectomy or radiation therapy. Surgery is preferred in younger women to avoid radiation exposure to the ovaries 85% 5-year survival, Stage II = Invades beyond uterus but not pelvic wall, 60% 5-year survival, Stage III = Involves pelvic wall or lower 1/3 of vagina 33% 5-year survival, Stage IV = Bladder or rectal invasion, extension beyond true pelvis 7% 5-year survival.

Stages II-IV usually are treated with radiation therapy, which may be combined modalities and often includes brachytherapy. For stage II to IV chemotherapy is usually given; this acts as a radiosensitizer [3].

10.2.2 Current approach to 1 diagnosis & staging

Papanicolaou SmearScreening Recommendations The Papanicolaou smear is 90-95% accurate for detection of early lesions such as CIN, but is less sensitive for detecting invasive cancers or fungating masses. False positive results can occur with inflammation, necrosis and hemorrhage. The American Cancer Society recommends 2 consecutive annual Papanicolaou smears starting at age 20 or with onset of sexual activity. If both are negative, then triannual screening is recommended. Atypical findings or low-grade CIN require follow-up every 3 to 6 months, plus testing for HPV [3]. Biopsy Colposcopically directed biopsy is necessary for any visible lesion irrespective of the Pap smear finding or for high-grade CIN. The reliability of biopsy has been critically discussed by Mountford et al. [34], who state: accurate diagnosis of cervical cancer currently

166

relies on the pathology of the tissue obtained during colposcopic examination and this process is not free of problems. Sampling errors can be introduced at the biopsy stage to be compounded by processing artefacts in the laboratory and subjective assessment of the section by the pathologist. By definition, the in situ phase of an epithelial neoplasm contains cells which are morphologically indistinguishable from those in the invasive state. Diagnosis of invasive cancer therefore rests, not on cytological criteria, but on histological evidence of destructive invasion (pp. 1522-1523). MRI MRI has been considered the imaging method of choice for staging cervical cancer. The presence of a low signal intensity stripe of peripheral cervical stroma on MRI is 95% specific for excluding parametrial invasion. This is important for recognizing lesions that could be surgically resected. MRI can also identify vaginal invasion and other manifestations of advanced disease [18]. MRI is helpful in distinguishing persistent or recurrent carcinoma from fibrosis or necrosis. Fibrosis usually appears as an area of low signal intensity on both T1 and T2 weighted images, whereas tumor is seen as moderate to high intensity on T2 weighted images. However, inflammation, edema, necrosis and capillary hypervascularity associated with radiation therapy can also show high intensity on T2 weighted images. In general, MRI is non-specific for 3 to 6 months after radiation therapy [35].

10.2.3 In vivo proton MRS in cervical cancer diagnostics

Lee et al. [36] were the first to publish a report on in vivo MRS7 of lesions of the uterine cervix. Their study included 51 patients with cervical cancer and 3 healthy volunteers. Resonances at 0.9, 1.3 ppm (triglycerides), 3.0 ppm (creatine) and 3.25 ppm (choline) were seen in cervical cancer, but not in normal cervix. The triglyceride peak was seen in 39 of 44 cases of squamous cell carcinoma and in 1 or 7 cases of adenocarcinoma. In the normal cervix, no peaks at 0.9 to 1.3 ppm were seen at TE = 135ms, and very small peaks at 0.9 to 1.3 ppm with 16x magnification. Choline (3.2 ppm) was seen in all the cancers. At TE = 20ms, normal lipids were seen abundantly. A summary of the findings in squamous cell and adenocarcinoma is presented in Table 10.4. The authors [36] suggest that the triglyceride peak at 1.3 ppm

Lee et al. [36] used 1.5T, body coil for transmitting the signal and endovaginal surface coil for receiving, single voxel, TR/TE: 3000/20 and 135 ms, PRESS, assessed presence or absence of peaks.

167

may be helpful for diagnosing squamous cell carcinoma, while the peak at 2.0 ppm might be specific for adenocarcinoma. Table 10.4 Presence of Metabolites in Cervical Cancer using in vivo MRS
(From data of Lee et al. [36])
Triglycerides (1.3 ppm) Creatine (3.0 ppm) Choline (3.2 ppm) NAcetylneuraminic Acid (2.0 ppm) None of 44 patients

Squamous cell Carcinoma

39 of 44 patients

44 of 44 patients

44 of 44 patients

Adenocarcinoma

3 of 7 patients

7 of 7 patients

7 of 7 patients

6 of 7 patients

More recently, Allen et al. [35] compared 8 healthy volunteers and 26 patients with histologically proven cervical carcinoma (16 patients scanned prior to RT, 10 patients at follow-up post RT, and 8 patients with persistent or suspected recurrent disease), using proton MRS8. All 16 patients prior to RT had choline at 3.25 ppm. Some had -glutamyl at 1.8 to 2.4 ppm, consistent with glutamate, glutamine or glutathione. Choline was also present in retained parametria and sidewall tissues of a patient with a cut-through hysterectomy. Proton MRS used in follow-up after radiation therapy Allen et al. [35] also used in vivo proton MRS for follow-up after radiation therapy. Of the 10 patients post-RT without clinical evidence of persistence or recurrence8 patients had spectra similar to the healthy volunteers with no choline resonance. Two patients had a small persistent choline resonance, which resolved at 6 months in one case, while the second patient is being re-evaluated. Of the 8 patients with unequivocal (2 patients) or suspected (6 patients) recurrence, all had choline appearing. One of these patients had a normal pelvic exam and MRI, but atypia on biopsy and seems to have had choline. At 56 months she was without evidence of disease, including having a normal Pap smear. Another patient suspected of having recurrent disease had a small choline resonance, and large lipid and other resonances suggesting necrosis. She had a recurrence at 57 months and died 3 months later.
8

Allen et al. [35] used 1.5T, single voxel PRESS, TR/TE=1600/140ms, peaks identified by automated peak-finding algorithm.

168

10.2.4 In vitro MRS findings in cervical cancer diagnostics

In vitro proton MRS has also been used to analyze the metabolic characteristics of cervical cancer. Mountford et al. [37] reported a narrow lined lipid from squamous carcinoma of the cervix, whereas inflammatory cells generated a broad component at 1.3 ppm with a T2 relaxation < 350ms. Most specimens with dysplastic cells or HPV infection showed a lipid spectrum similar to the malignant spectrum. Some, but not all of the cancerous tissue showed long T2 at 1.2 and 1.3 ppm. The authors from Ref. [37] noted that small tissue samples (6 mm3) such as punch biopsy could be analysed and emphasized the possibility of detecting precursor states for cervical cancer. A few years later, a paper was published by Delikatny et al. [38] in which an 8.5T magnet was used to analyze 159 cervical biopsy specimens. These were compared with histopathology. A highresolution lipid was seen in 3 of 40 invasive carcinomas. The 119 preinvasive samples showed little or no lipid, but a strong unresolved resonance between 3.8 and 4.2 ppm. Peak ratios of methylene/methyl and unresolved/methylene accurately distinguished invasive from preinvasive epithelium (p <0.0001). The authors concluded that proton MRS could independently provide distinction between invasive and pre-invasive cervical lesions and thereby help in the clinical management. More recently, Mountford et al. [34] examined lipid chemistries of biopsies from uterine cervix as well as other cancers with chemical shift imaging. They compared 40 specimens of invasive carcinoma with 119 pre-invasive specimens. In 39 of 40 of the invasive specimens, an intense resonance was found at 1.3 ppm from methylene protons of acyl chains in mobile neutral lipid, but there were also contributions from methyl protons from lactate and threonine. Highgrade dysplastic specimens also showed resonances at 0.9, 3.2 and 5.2 ppm, and at 1.7, 2.0 and 3.0, but pre-invasive specimens lacked the intense methylene resonance at 1.3 ppm (p. 1523). The pre-invasive specimens also showed a broad featureless resonance between 3.4 and 4.2 ppm, this is termed the CH resonance. Plotting CH2/CH3 versus CH/CH2 allowed distinction between invasive and pre-invasive specimens with a sensitivity of 94% and specificity of 98%. Very recently, HRMAS was been applied to 8 cervical cancer tissue specimens and 8 from non-malignant cervical lesions [39]. Accurate classification was achieved, with high level of choline and amino acid residues, and lower levels of glucose distinguishing the malignant specimens.

169

10.3 Endometrial Cancer

10.3.1 Overview of epidemiological and clinical aspects
Incidence and prevalence/morbidity and mortality Endometrial cancer accounts for about 3.8% of all cancers in women worldwide. The highest rates are reported in North America and Eastern Europe, with lower incidence seen in Asia. There are approximately 34 000 cases diagnosed annually and 6 000 women die each year from endometrial cancer in the U.S. [3, 40]. Incidence rates rise steadily 5 to 10 years prior to menopause and peak at age 65-70. Except in Japan where the death rates have doubled since 1970, especially among younger women, worldwide there is a decreasing mortality rate due to endometrial cancer. Early detection is considered the likely reason for improved survival [3, 40]. Etiology/risk factors Gynecologically related risk factors
Chronic estrogen exposure Cited as the best-recognized risk factor. This exposure may be exogenous (estrogen-only HRT) or endogenous, due to low parity, polycystic ovary syndrome, estrogen-secreting tumors, early menarche or late menopause [3, 41].

Atypical adenomatous hyperplasia A likely pre-malignant stage, characterized by increased number of glands, cytological atypia. An estimated one-third progress to carcinoma [41].
Adenomatous endometrial polyps Patients with atypical polyps taking tamoxifen have a markedly increased incidence of atypical hyperplasia compared to those with normal endometrium [42] (see later in this subsection).

Inherited susceptibility:
Hereditary non-polyposis colon cancer A cited hereditary risk factor [43].

Lifestyle-related factors
Obesity Leads to conversion of androstendione from the adrenal to estrone. Diet could be important in light of the low prevalence in Asian countries where fat intake is low [44].

170

Medical Conditions and Related Exposures

Tamoxifen While an estrogen antagonist, tamoxifen also has agonist properties. It is somewhat difficult to separate out from the risk of endometrial carcinoma associated with breast cancer [41]. Tamoxifen increases the risk of endometrial cancer 1.3 to 7.5-fold; the risk rises with duration and cumulative dose. Endometrial cancers associated with tamoxifen are usually aggressive, highgrade types [45]. Prior radiation therapy to the pelvis

Diabetes mellitus, hypertension Cited as independent risk factors for endometrial cancer [41, 44].

Occupational and Environmental Factors There is very little published data concerning occupational and environmental factors and risk of endometrial cancers. A Finnish registry study of endometrial cancer linked to occupational titles from the 1970 census (2 833 cases) reports a relative risk of 1.2 for exposure to animal dust and 1.3 for sedentary work [46]. Adjustment for obesity was not noted. According to Ahlborg et al. [47] there is limited empirical support for associations between organochlorine exposure and endometrial cancer (the hypothesis being estrogen over-activity) looking also at in vitro and animal data. Clinical Presentation The most common manifestations of endometrial cancer are abnormal vaginal discharge or bleeding. Early endometrial cancer may be asymptomatic. Differential Diagnosis Relevant differential diagnoses of non-ovulatory vaginal bleeding include, inter alia:
Endometrial polyps Estrogen breakthrough bleeding (continuous estrogen stimulation) Uterine fibroids Bleeding disorders Idiopathic menorrhagia, peri-menopausal bleeding, post-menopausal endometrial atrophy [45, 48].

171

Grading and Staging Histopathology and grading About 75 to 80% of endometrial carcinomas are adenocarcinomas. Other histopathologic types include mucinous (5%), papillary serous carcinomas, clear cell, ciliate and secretory types [3]. Uterine sarcomas are a rare type of cancer affecting the uterus. Grading is as follows [3, 41]:
Grade I = Highly differentiated (identifiable endometrial glands) Grade II = Some solid areas Grade III = Largely solid and undifferentiated

Staging
Stage IA = Confined to endometrium Stage IB = Invades < of myometrial thickness Stage IC = Invades > of myometrial thickness IIA = Endocervical glands only IIB = Cervical stroma IIIA = Uterine serosa or adnexa, or positive peritoneal cytology IIIB = Vaginal metastases IIIC = Spread to pelvic or paraaortic lymph nodes IVA = Invades bladder or rectosigmoid mucosa IVB = Distant metastases.

Treatment and Prognosis Stage I is treated with total abdominal hysterectomy and bilateral salpingo-oophorectomy. More advanced stages and poor histological grade are treated with surgery and radiation therapy. Stage IV is handled palliatively. Chemotherapy is generally not very successful, although some patients respond to doxorubicin and cisplatin [3]. Stage I disease has about a 90% 5-year survival. For Stage II about 80% of patients survive five or more years. Survival is much worse for Stage 3 (an estimated 30% 5-year survival) and only about 9% of patients with Stage IV disease live longer than 5 years after diagnosis [3].

172

10.3.2 Current approach to 1 diagnosis & staging

Transvaginal ultrasound TVS is usually the first imaging examination for women with dysfunctional uterine bleeding [44]. High-resolution endovaginal probes allow visualization of endometrium96% of patients with endometrial cancer have endometrial thickness > 5mm. TVS can identify patients at low risk for endometrial disease, obviating the need for endometrial sampling in this sub-group. The cut-off value of 5 mm is highly sensitive, but has lower specificity, leading to unnecessary sampling procedures performed in order to avoid missing important endometrial pathology [45].
Diagnostic Accuracy for women taking HRT Significantly higher false-positive rate (specificity 77% (95% CI = 75 79%) versus 92% (95% CI = 90 94%) for those not taking HRT). This is not surprising, since HRT leads to increased endometrial thickness [45]. Hysterosonography (HSG) The value of HSG is emphasized by Reinhold and Khalili [45], who consider it to be one of the minimally invasive procedures that plays an important role in the detection and characterization of endometrial pathology (p. 531). HSG is more accurate than endovaginal sonography for the detection, localization, and characterization of endometrial pathology. Whereas endovaginal sonography shows only abnormal thickening of the endometrium, HSG can more precisely detect endometrial polyps and submucosal myomas with an intracavitary component, and thereby can select patients who benefit most from hysteroscopic-guided removal. Whereas endometrial carcinoma on TVS shows only thickening, HSG shows an irregular broad-based mass. HSG can distinguish true from apparent endometrial thickening as seen on TVS. While the exact role of HSG has not yet been defined, proposed guidelines for its implementation are for patients with: Endometrial thickening on TVS and negative biopsy results Indeterminate findings on TVS Persistent bleeding and negative TVS or endometrial biopsy [45]. Doppler Ultrasound Doppler indices and color Doppler vascularity of the endometrium have been used to distinguish benign from malignant endometrial pathology. There are various threshold values for resistive index. Increased vascularity is generally associated with malignant endometrium [45].

173

Endometrial biopsy or fractional dilatation and curettage (D & C) These are the usual procedures for diagnosis. However, they can be considered definitive only when positive [3].
Endometrial biopsy About 15% of endometrial biopsies provide an inadequate sample for histopathologic assessment. There are reports of false negatives of over 65%, although the sensitivity is usually found to be 85% for endometrial carcinoma. Endovaginal sonography together with biopsy is recommended, especially to find polyps and fibroid myomas [45]. According to Burke et al. [41] a correctly performed endometrial biopsy includes an adequate amount of tissue obtained from multiple passes through the uterus has a diagnostic accuracy equivalent to that of surgical curettage under anaesthesia (pp. 15751576).

D & C Difficult to assess the sensitivity and specificity. In patients whose uteri were subsequently removed, up to 10% of endometrial lesions had been missed with D & C. False negative rates reported to be 2 6 %. In 60% of patients less than half of the endometrium was sampled. As an invasive procedure, D & C is associated with complications including bleeding, infection and uterine perforation [45].

MRI MRI is generally considered the preferred imaging mode for pretreatment evaluation of endometrial cancer [44, 50], with contrastenhanced MR imaging showing the highest efficacy [51]. Dynamic contrast-enhance MRI with delayed, fat-suppressed T1 weighted images is found to be helpful for detecting and characterizing endometrial pathology [45].
Normal uterus on MRI On T2 weighted MRI, the endometrium appears as a central hyper-intense structure, surrounded by the hypo-intense junctional zone, which is the inner layer of the myometrium. The endometrium is usually iso-intense with T1 weighting [45]. Non-cancerous uterine pathologic on MRI Endometrial hyperplasia appears as diffuse or, less frequently, localized thickening of the endometrial complex, with well-defined borders. MRI cannot distinguish between hyperplasia with and without cellular atypia. Polyps appear as intermediate signal intensity on T1 weighted images. They are either slightly hypo-intense or iso-intense on T2 weighted images.

174

Fibroids are sharply marginated low intensity on T2 weighted images, and appear hypovascular after gadolinium. With degeneration, they can be of mixed signal intensity [45]. Endometrial cancer on MRI Endometrial carcinoma generally appears with an intensity intermediate between that of the normal endometrium and the myometrium. Stage IA endometrial carcinoma shows a thickened endometrial stripe and diffuse or focal abnormal signal intensity. This abnormal tumor signal intensity extends progressively into more of the myometrium with Stages IB and IC. With myometrial invasion, the tumor borders are frequently irregular or ill defined [44, 45, 49]. MRI thereby shows the depth of myometrial invasion, an important prognostic indicator, with an accuracy of 75-95%. MRI can also show extension of endometrial carcinoma into the endocervix. However, MRI staging accuracy is diminished when uterine zonal architecture has been distorted, e.g. by fibroids or when the junctional zone is not present, as can be the case among women after menopause. Moreover, in very early stages of endometrial carcinoma, the MRI findings are non-specific [49].

Serum tumor markers: CA-125, CA-19-9 CA-125 may indicate extrauterine spread of endometrial cancer, as well as myometrial invasion. CA 19-9 levels are often abnormal with endometrial cancer [41, 52]. FDG-PET FDG-PET has been useful in detecting recurrent endometrial cancer; in some cases these were asymptomatic. Whole-body scanning is reportedly helpful both for confirming and for ruling out disease recurrence [53]. Screening recommendations for endometrial carcinoma Currently, there is controversy about whether to screen asymptomatic women.
Data and arguments against screening asymptomatic women Comparing the clinico-pathologic findings in 21 asymptomatic patients to 427 symptomatic with endometrial cancer, Jobo et al. [54] found no relation to survival with univariate analysis, and no significant difference in grade, nodal or adnexal metastases, or several other criteria, though depth of invasion and histopathology differed. The authors do not recommend screening asymptomatic women. Robertson [43] states: routine screening for endometrial carcinoma is currently not justified (p. 657). The American Cancer Society and the American College of Obstetricians and Gynecologists

175

do not include annual TVS in their recommendations for asymptomatic women [20]. Data and arguments for screening asymptomatic women Ciatto et al. [55] followed 2 240 asymptomatic women after menopause as well as 1 220 women with abnormal uterine bleeding (AUB). For asymptomatic women the 4 mm cut-point showed 66.7% sensitivity and 92.1% specificity, positive predictive value of 2.2% and negative predictive value of 99.9%. There were 6 asymptomatic patients who developed endometrial carcinoma (EC) (mean thickness 4.67 range 0 10 mm) compared to 1.68 mm (range 0 20 mm) in those without EC. For those with AUB, the authors found that 4 mm on TVS showed a 91.1% sensitivity and 79.8% specificity with a positive predictive value of 14.8% and negative predictive value of 99.6% for developing endometrial carcinoma. A total of 45 patients with bleeding developed EC, with mean thickness of 8.0 mm (range 0 - 44 mm) versus 2.46 (range 0 - 20 mm) for those who did not develop EC. The authors conclude: the study confirms the usefulness of measuring endometrial thickness (half layer cut off = 4 mm) with transvaginal ultrasound in asymptomatic postmenopausal women to indicate further special surveillance and in subjects with AUB to indicate immediate invasive assessment (p. 437). Screening for women taking tamoxifen The American College of Radiology in 1996 recommended TVS as the first line imaging modality for patients taking tamoxifen [45]. The role of MRI in screening Currently, MRI is not considered to have a role in screening for endometrial pathology, due inter alia, to the lack of specificity for early endometrial cancer [45, 49].

10.3.3 In vitro MRS findings in endometrial cancer diagnostics

With respect to endometrial cancer, there are no published studies to the best of our knowledge, in which in vivo MRS has been applied. There are only very minimal in vitro MRS data available on endometrial cancer. Among 100 consecutive cases of patients with endometrial carcinoma, prior to surgery, Schneider et al. [52] reported that in vitro MRS analysis of pre-surgical curettage specimens predicted stage in 84.6% of cases. Spectroscopic details are not given in their paper, however.

176

HRMAS was used to metabolically characterize Ishikawa cells (a human cell line derived from endometrial adenocarcinoma) in a study by Griffin et al. [56]. The obtained spectrum showed nucleotide derivatives of uridine and adenosine. Cells exposed to tamoxifen showed increased ethanolamine (3.26 ppm), glucose (3.34 3.94 ppm), glutamate (2.14, 2.32 ppm), tyrosine (7.24 ppm) uridine (7.85 ppm) and adenosine (8.20 ppm), with a relative decrease in myoinositol (3.30, 3.62, 3.55 ppm). The nucleotide changes seem to the authors to suggest that tamoxifen affects RNA transcription, while the changes in ethanolamine and myoinositol concentrations appear to indicate cell membrane turnover [56].

10.4 Comment on difference in approach to gender-specific cancers using MRS and MRSI
This author is struck by the wealth of studies using in vivo MRS and MRSI for prostate cancer with noteworthy improvements in diagnostic accuracy (see Chapter 9), and the dearth, and in the case of endometrial cancer total absence, of publications applying MRS and particularly MRSI to gynecological cancers, as reviewed in the present chapter. We are hopeful that the benefits of molecular imaging through magnetic resonance will soon be made available with the aim of further improving diagnostics of these gynecologic cancers.

References
[1] R.F. Ozols, P.E. Schwartz, P.J. Eifel, Ovarian cancer, fallopian tube carcinoma and peritoneal carcinoma, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 1597-1632. [2] I.B. Runnebaum, E. Stickeler, Epidemiological and molecular aspects of ovarian cancer, J. Cancer Res. Clin. Oncol. 127, 73-79 (2001). [3] R.C. Young, Gynecologic Malignancies, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 620-625. [4] D. Gertig, D. Hunter, Ovarian cancer, in: H-O. Adami, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, Oxford, 2002, p. 378-399. [5] S.A. Funt, L.E. Hann, Detection and characterization of adnexal masses, Radiol. Clin. N. Am. 40, 591-608 (2002). [6] A. Venn, D. Healy, R. McLachlan, Cancer risks associated with the diagnosis of infertility, Best Pract. Res. Clin. Obstet. Gyn. 17, 343-367 (2003).

177

[7] T.L. Marx, A.E. Mehta, Polycystic ovary syndrome; pathogenesis and treatment over the short and long term, Cleveland. Clin. J. Med. 70, 31-33, 36-41 (2003). [8] K. Vasama-Neuvonen, E. Pukkala, H. Paakkulainen, P. Mutanen, E. Weiderpass, P. Boffetta, N. Shen, T. Kauppinen, H. Vainio, T. Partanen, Ovarian cancer and occupational exposures in Finland, Am.J. Indust. Med. 36, 83-89 (1999). [9] N. Shen, E. Weiderpass, A. Antilla, M.S. Goldberg, K. Vasama-Neuvonen, P. Boffetta, H. Vainio, T. Partanen, Epidemiology of occupational and environmental risk factors related to ovarian cancer, Scand J. Work Environ Health 24, 175-182 (1998). [10] S.A. Petralia, M. Desemeci, E.E. Adams, S.H. Zahm, Cancer mortality among women employed in health care occupations in 24 U.S. states, 1984-1993, Am. J. Indust. Med. 36, 159-165 (1999). [11] S. Koifman, R.J. Koifman, A. Meyer, Human reproductive system disturbances and pesticide exposure in Brazil, Cad. Saude Publica 18, 435-445 (2002). [12] J. Dich, S.H. Zahm, A. Hanberg, H.O. Adami, Pesticides and cancer, Cancer Causes & Control 8, 420-443 (1997). [13] T. Shields, G. Gridley, T. Moradi, J. Adami, N. Plato, M. Dosemeci, Occupational exposures and the risk of ovarian cancer in Sweden, Am. J. Indust. Med. 42, 200-213 (2002). [14] H. Langseth, A. Andersen, Cancer incidence among women in the Norwegian pulp and paper industry, Am J. Indust. Med. 36, 108-113 (1999). [15] M.A. Bulbulyan, S. A. Ilychova, S.H. Zahm, S.V. Astashevsky, D.G. Zaridge, Cancer mortality among women in the Russian printing industry, Am. J. Indust. Med. 36, 166-171 (1999). [16] D.S. Heller, R.E. Gordon, C. Westhoff, S. Gerber, Asbestos exposure and ovarian fiber burden, Am. J. Indust. Med. 29, 435-439 (1996). [17] F.V. Coakley, Staging ovarian cancer: role of imaging, Radiol. Clin. N.Am. 40, 609-636 (2002). [18] R.R. Barakat, H. Hricak, What do we expect from imaging? Radiol. Clin. N. Am. 40, 521-526 (2002). [19] A.K. Sood, J.I. Sorosky, M.S. Gelder, R.E. Buller, B. Anderson, E.J. Wilkinson, J.A. Benda, L.S. Morgan, Primary ovarian sarcoma Analysis of prognostic variables and the role of surgical cytoreduction, Cancer 82, 1731-1737 (1998). [20] L. Cohen, Should transvaginal ultrasound be performed at annual examination in asymptomatic women? Int. J. Fertil. 48, 150-153 (2003) [21] S.W. Cho, S.G. Cho, J.H. Lee, H-J Kim, M.H. Lim, J.H. Kim, C.H. Suh, In-vivo proton magnetic resonance spectroscopy in adnexal lesions, Korean J. Radiol. 3, 105-112 (2002). [22] C. Anderiesz, M.A. Quinn, Screening for ovarian cancer, Med. J. Aust.178, 655-656 (2003). [23] F. Alexander-Sefre, U. Menon, I.J. Jacobs, Ovarian cancer screening, Hospital Med. 63, 210213 (2002). [24] P.J. Paley, Ovarian cancer screening; are we making any progress? Curr. Opin. Oncol. 13, 399-402 (2001). [25] K. Togashi, MR imaging of the ovaries: normal appearance and benign disease, Radiol. Clin. N. Am. 41, 799-811 (2003). [26] G.E. Sarty, E.J. Kendall, J. Loewy, A. Dhir, O.A. Olatunbosun, R.A. Pierson, Magnetic resonance diffusion imaging of ovarian masses: a first experience with 12 cases, MAGMA 16, 182193 (2004).

178

[27] T. Okada, M. Harada, K. Matsuzaki, H. Nishitani, T. Aono, Evaluation of female intrapelvic tumors by clinical proton MR spectroscopy, J. Magn. Reson. Imaging 13, 912-917 (2001). [28] I.C. Smith, D.E. Blandford, Diagnosis of cancer in humans by 1H NMR of tissue biopsies, Biochem. Cell Biol. 76, 472-476 (1998). [29] J.C. Wallace, G.P. Raaphorst, R.L. Somorjai, C.E. Ng, M. Fung Kee Fung, M. Senterman, I.C. Smith, Classification of 1H MR spectra of biopsies from untreated and recurrent ovarian cancer using linear discriminant analysis, Magn. Reson. Med. 38, 569-576 (1997). [30] E.A. Boss, S.H. Moolenaar, L.F.A.G. Massuger, H. Boonstra, U.F.H. Engelke, J.G.N. de Jong, R.A. Wevers, High-resolution proton nuclear magnetic resonance spectroscopy of ovarian cyst fluid, NMR Biomed. 13, 297-305 (2000). [31] L.F.A.G. Massuger, P.B.J. van Vierzen, U. Engelke, et al., 1H-magnetic resonance spectroscopy. A new technique to discriminate benign from malignant ovarian tumors, Cancer 82, 1726-1730 (1998). [32]. S. Stuver, H-O. Adami, in: H-O. Adami, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, Oxford, 2002, p. 340-358. [33] D.A. Savitz, K.W. Andrews, L.A. Brinton, Occupation and cervical cancer, J. Occup. Environ. Med. 37, 357-361 (1995). [34] C.E. Mountford, W.B. Mackinnon, P. Russell, A. Rutter, E.J. Delikatny, Human cancers detected by proton MRS and chemical shift imaging ex vivo, Anticancer Res. 16, 1521-1532 (1996). [35] J.R. Allen, R.W. Prost, O.W. Griffith, S.J. Erickson, B.A. Erickson, In vivo proton (H1) magnetic resonance spectroscopy for cervical carcinoma, Am. J. Clin. Oncol. 24, 522-529 (2001). [36] J.H. Lee, K.S. Cho, Y.M. Kim, S.T. Kim, C.W. Mun, J.H. Na, J.E. Mok, T.H. Lim, Localized in vivo 1H nuclear MR spectroscopy for evaluation of human uterine cervical carcinoma. Am. J. Roentgenol. 170, 1279-1282 (1998). [37] C.E. Mountford, E.J. Delikatny, M. Dyne, K.T. Holmes, W.B. Mackinnon, R. Ford, J.C. Hunter, I.D. Truskett, P. Russell, Uterine cervical punch biopsy specimens can be analyzed by 1H MRS, Magn.Reson. Med. 13, 324-331 (1990). [38] E.J. Delikatny, P. Russell, J.C. Hunter, R. Hancock, K.H. Atkinson, C. van Haaften-Day, C.E. Mountford, Proton MR and human cervical neoplasia: ex vivo spectroscopy allows distinction of invasive carcinoma of the cervix from carcinoma in situ and other preinvasive lesions, Radiology 188, 791-796 (1993). [39] B. Sitter, T. Bathen, B. Hagen, C. Arentz, F.E. Skjeldestad, I.S. Gribbestad, Cervical cancer tissue characterized by high-resolution magic angle spinning MR spectroscopy, MAGMA 16, 174181 (2004). [40] I. Persson, H-O. Adami, Endometrial Cancer, in: H-O. Adami, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, Oxford, 2002, p. p. 359-377. [41] T.W. Burke, P.J. Eifel, F.M. Muggia, Cancers of the uterine body, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 1573-1594. [42] S.R. Goldstein, Controversy about uterine effects and safety of SERMs: the sage continues, Menopause 9, 381-4 (2002). [43] G. Robertson, Screening for endometrial cancer, Med. J. Australia 178, 657-659 (2003). [44] S.M. Ascher, C. Reinhold, Imaging of cancer of the endometrium, Radiol. Clin. N. Am. 40, 563-576 (2002).

179

[45] C. Reinhold, I. Khalili, Postmenopausal bleeding: value of imaging, Radiol. Clin. N. Am. 40, 527-562 (2002) [46] E. Weiderpass, E. Pukkala, K. Vasama-Neuvonen, T. Kauppinen, H. Vainio, H. Paakkulainen, P. Boffeta, T. Partanen, Occupational exposures and cancers of the endometrium and cervix uteri in Finland, Am. J. Indust. Med. 39, 572-580 (2001). [47] U.G. Ahlborg, L. Lipworth, L. Titus-Ernstoff, C.C. Hsieh, A. Hanberg, J. Baron, Trichopoulos, H.O. Adami, Organochlorine compounds in relation to breast cancer, endometrial cancer and endometriosis: an assessment of the biological and epidemiological evidence, Crit. Rev. Toxicol. 25, 463-531 (1995). [48] M. Mints, Idiopathic menorrhagia, Studies of angiogenesis and surgical therapy, Doctoral dissertation, Karolinska Institutet, Stockholm 2003. [49] A.E. Li, D.A. Bluemke, Magnetic resonance imaging in: de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. p. 669-679. [50] S. Taieb, L. Ceugnart, E. Leblanc, A. Chevalier, V. Cabaret, D. Querleu, MR imaging of endometrial carcinoma: role and limits, Bull. Cancer 89, 963-968 (2002). [51] K. Kinkel, Y. Kaji, K.K. Yu, M.R. Segal, Y. Lu, C.B. Powell, H. Hricak, Radiologic staging in patients with endometrial cancer: a meta-analysis, Radiology 212, 711-718 (1999). [52] J. Schneider, M. Centeno, F. Sez, J. Genoll, A. Ruibal, Preoperative CA-125, CA 19-9 and nuclear magnetic resonance in endometrial carcinoma: correlation with surgical stage, Tumor Biol. 20, 25-29 (1999). [53] T. Belhocine, C. De Barsy, R. Hustinx, J. Willems-Foidart, Usefulness of (18) F-FDG PET in the post-therapy surveillance of endometrial carcinoma. Eur. J. Nucl. Med. Molecular Imaging 29, 1132-1139 (2002). [54] T. Jobo, T. Arai, R. Sato, H. Kuramoto, Clinicopathologic relevance of asymptomatic endometrial carcinoma, Acta Cytologica 47, 611-615 (2003). [55] S. Ciatto, S. Cecchini, G. Gervasi, A. Landini, M. Zappa, E. Crocetti, Association of endometrial thickness assessed at trans-vaginal ultrasonography to endometrial cancer in postmenopausal women asymptomatic or with abnormal uterine bleeding, Radiol. Med. 104, 437442 (2002). [56] J.L. Griffin, J.C.M. Pole, J.K. Nicholson, P.L. Carmichael, Cellular environment of metabolites and a metabolic study of tamoxifen endometrial cells using gradient high resolution magic angle spinning 1H NMR spectroscopy, Biochim. Biophys. Acta. 1619, 151-158 (2003).

181

Chapter 11

Head and Neck Cancers

_______________________________________________________________________________

Carcinomas of the head and neck arise from mucosal surfaces and are usually of the squamous cell type. These include cancers of the oral cavity, paranasal sinuses, nasopharynx, oropharynx, hypopharynx and larynx [1].

11.1 Overview of epidemiological & clinical aspects

Incidence and prevalence/morbidity and mortality The worldwide incidence of head and neck tumors is estimated to be greater than 500 000 new cases per year. In Europe and North America head and neck tumors generally are from the oral cavity, oropharynx or larynx. Head and neck malignancies account for approximately 5% of adult cancers in the U.S. In the Mediterranean and Far East regions, nasopharyngeal cancer occurs more often [1]. Oral Cancers Oral cancers account for about 5% of malignancies worldwide among men, and 2.5% among women. In the year 2000, there were an estimated 207 000 deaths from oral cancer. Mortality rates are high both in the developed and the developing world, with rates increasing in Southern and Eastern Europe, as well as in Japan and among African-Americans. The incidence rates increase with age, but there are some exceptions (e.g. Bas-Rhin, France) [2].

182

Most importantly is that the morbidity and mortality associated with oral cancer are largely preventable. First, precancerous lesions precede the majority of tumors, and treatment leads to remission of the pre-cancer and a decreased risk of malignant transformation Oral cancer exists in a preclinical detectable stage, and detection involves a simple examination and palpation of the oral cavity With treatment at an early stage, the prognosis is good (pp. 118-119)[2]. Nasopharyngeal Cancers The risk is usually low except in Hong Kong, Taipei, among Arctic Eskimos, many indigenous people of Southeast Asia and Arab populations of North Africa and Kuwait [2].

Etiology/risk factors Epidemiological investigation of head and neck cancers is difficult because of the many anatomic sub-sites. Some risk factors are important for many sub-sites, while others contribute only to some [2]. Risk factors for Oral Cancers Genetic and Molecular Epidemiology Familial susceptibility is recognized. This appears to involve common genetic changes that together with environmental exposures can lead to malignancy. This gene-environmental interaction most likely involves genes that encode enzymes for metabolism of tobacco, alcohol and dietary factors [2].
High Penetrance Gene Mutations

In India this familial aggregation appears to be conferred by autosomal dominant inheritance. Inherited defects in p53 are associated with increased familial incidence of many cancers including oral carcinomas. Inherited defective DNA repair mechanisms, e.g. xeroderma pigmentosum, are associated with increased risk for oral cancers [2]. Low Penetrance Gene Mutations Polymorphisms in genes may possibly play a role: Glutathione S-Transferase detoxifies polyaromatic hydrocarbons (PAH) such as benzo(a) pyrene in tobacco smoke; the exact role is still unclear in the etiology of oral cancer.

183

Cytochrome P450 enzymes that catabolize PAHs including benzo(a) pyrene and other constituents of tobacco smoke may be involved, as per evidence from China and France. Alcohol Dehydrogenase (ADH3) fast metabolism of alcohol to acetaldehyde which is carcinogenic, expressed in the oral cavity; there is some evidence that the fast metabolizing allele may interact with alcohol consumption [2].

Somatic Mutations The development of oral cancer involves several molecular steps, with somatic mutations in tumor-suppressor genes and oncogenes. Mutations in p53 have been shown in over 60% of oral cancers, as well as in pre-cancerous lesions [2]. The p53 mutation is an early event in head and neck tumors [3]. Changes in p16, which normally blocks the progression of cell growth cycle at G1, and loss of p53 are seen early in oral carcinogenesis, and lead to unregulated cell growth. Altered p16 is associated with tobacco use and betel quid (chewed tobacco) [2].

As stated by Sidransky [4] states: identification of the critical genetic changes that drive the neoplastic process has provided a preliminary progression model for head and neck cancer (p. 789). Exposures related to lifestyle
Tobacco Tobacco is an established risk factor for oral cancers. In the U.S. 80% of oral cancers have been attributed to the use of tobacco [5]. Smoking cigarettes exposes the oral cavity to a number of carcinogens: PAHs, aldehydes and nitrosamines. The association is strong and consistent, e.g. an 11-fold increase is reported among cigarette smokers in Italy compared to non-smokers. This is a dose-dependent relationship; quitting smoking leads to a dramatic decrease in risk [2]. Smokeless tobacco contains N-nitrosamines; chewing tobacco and snuff are associated with high risk of oral cancer, especially of the cheek and gum. The high rates of oral cancer in India and Southeast Asia are strongly related to use of smokeless tobacco. The location within the oral cavity is quite specific to the site and type of product used [2]. Alcohol Epidemiologic data collected over the past 30 years provide consistent evidence that alcohol elevates the risk of oral cancer, both alone and synergistically with tobacco (p. 125) [2]. Diet Low consumption of fruits and vegetables Carotenoids may be protective [1]

184

Combined Effects (Tobacco plus alcohol) There is a consistently observed interaction between tobacco and alcohol use in the risk of oral cancer [2, 6]. The mucosal surface of the entire pharynx is exposed to alcohol and tobacco-related carcinogens, and is at risk for the development of a pre-malignant or malignant lesion that can progress to invasive carcinoma. Alternatively, multiple synchronous or metachronous1 cancers can develop. In fact, patients with early-stage head and neck cancer are at greater risk of dying of a second malignancy than of dying from a recurrence of the primary disease (p. 560) [1].

Infections
Human Papillomavirus HPV has been implicated, based upon several lines of evidence: Women with cancer of the cervix closely associated with HPV infection, have a 2-fold increased risk of subsequent oral carcinoma. There is a 50% prevalence of HPV in oral cancer patients. The role of HPV in oral cancer may be to inactivate tumor-suppressor genes. Herpes Simplex Virus (HSV) HSV has been associated with increased risk, especially of cancer of the lip. Chronic syphilis is associated with risk of cancer of the tongue [2].

Occupation According to Mucci and Adami [2], work-related factors do not contribute to a large proportion of the total oral cancers. They note that the cohort studies generally lack data on confounders such as smoking.
Aromatic amines and to phenoxy herbicides Reported to be associated with increased risk of head and neck cancers. Rubber workers Those exposed to high levels of nitrosamines (especially in the high-risk processing departments) show increased risk (3-4 fold) after 1 or more years of this work. Cooks Reported to have over 7-fold risk in cohort studies. They are exposed to volatile carcinogenic compounds, but also may use alcohol and tobacco, and are at risk for hepatic cirrhosis [2].

1 Metachronous: abnormal tissue growth that arises from an independent oncogenic event, and may or may not be related to the primary neoplasm, although underlying risk factors may cause both cancers.

185

Risk factors for Nasopharyngeal Cancers Salted fish and Preserved Foods These foods contain high levels of N-nitroso compounds. The relative risk of a diet high in salted fish ranges from 2.1 to 37.7. This is common in boat dwellers of Hong Kong, as well as Taiwan and Southeast Asia. Ingestion of preserved foods, especially during in the early years of life, is associated with increased risk of nasopharyngeal cancer, and may, at least in part, explain the elevated risk among people of Southeast Asia, Eskimos and Arab peoples of North Africa [2]. Epstein-Barr Virus (EBV) Antibody titers to EBV are higher among patients with nasopharyngeal cancer compared to referents. However, most adults are EBV antibody positive to at least one viral gene product [2]. HLA Polymorphisms2 The evidence here is limited, and difficult to acquire. However, genetic factors are considered likely to be of importance [2]. Occupation Environmental Exposures Occupations reported to be associated with elevated risk of nasopharyngeal cancer include agricultural workers, wood cutters / woodworkers, textile workers, fishermen, metal smelting, blacksmiths, cooks, machine-tool operators, and other specific manufacturing occupations [1, 7-11].
Nickel Found in high levels in the environment of high-risk areas for nasopharyngeal carcinoma in China. In vitro studies suggest an interaction between EBV and nickel in contributing to nasopharyngeal cancer [12]. Exposure to nickel is high in electroplating, welding as well as nickel refining. Nickel-induced carcinogenesis is likely to involve genetic and epigenetic3 routes. Cancers of the nose and nasal sinuses have been reported among workers exposed to nickel in cutlery factories, alkaline battery manufacture as well as nickel refining [13-14].

2 HLA is the acronym for the human leukocyte antigen complex, which is the human major histocompatibility complex. This is a region on chromosome 6, which is densely packed with expressed genes. The best known are HLA class I and II genes whose products are critical for immunological specificity and histocompatibility in transplantation. 3 Epigenetic: heritable alterations in gene expression but without changes in DNA sequence.

186

Clinical Presentation and Approach This depends upon the sub-site. A sore or difficulties in swallowing are frequent presentations [2]. The symptoms are often non-specific. Laryngeal cancer usually presents as sudden hoarseness. Hoarseness of longer than 3 weeks duration requires laryngoscopic examination. The most common clinical manifestation of oral cancer is an indurated ulcer. If an oral ulcer is present for over 3 weeks, biopsy should be performed [15]. Nasopharyngeal cancer is often asymptomatic, but may present as unilateral serous otitis media due to Eustachian tube obstruction, nasal obstruction or epistaxis. In the advanced stage, cranial neuropathies may be the presenting clinical manifestation [1]. Physical Signs It is important to scrutinize all visible mucosal surfaces and to palpate the floor of the mouth, the tongue and the neck. Leukoplakia (white mucosal patch) and erythroplakia (red mucosal patch) can be premalignant lesions or carcinoma in situ. All visible lesions should be biopsied, and referral made to an otorhinolaryngologist [1]. Diagnosis Importance of Surveillance / Early Detection These cancers are accessible, so that biopsy is made for a definitive histopathologic diagnosis. As emphasized by Schantz et al. [16] given the degree to which the oral cavity and upper aerodigestive tract can be easily examined, it would also seem that screening for head and neck cancers would be a readily accomplishable goal. Diminished mortality would be readily achievable. The significance of screening is emphasized in a review by Smart, who reported that 94% of head and neck cancer patients had seen a physician at least 1 year before diagnosis. Each patient reported an average of 11 physician visits within a 3-year period before diagnosis {The review [17]} emphasizes that with appropriate training and practice of systematic screening habits by examining physicians, head and neck cancer may be diagnosed considerably earlier (p. 799). It has been stated [17]: although the incidence and mortality rate of oral cancer is nearly double that of cancer of the cervix (30 000 versus 13 500 and 7 950 versus 400, respectively), conducting a pelvic examination and Pap smear appears more acceptable than looking in the mouth. The inspection of the oral cavity should be part of every physical examination in the dentists and physicians office. Ninety percent of all squamous cell cancers arise from the floor of the mouth,

187

the ventrolateral aspect of the tongue and the soft palate complex. The detection rate is increased from approximately 1 per 1000 in asymptomatic individuals older than 50 years to 1 in 200 in high-risk smokers and drinkers and to 1 in 7 for individuals once treated for oral cancer (p. 1061). Screening for second primary head and neck cancers is also of vital importance [16]. Molecular Assays As pointed out by Van Houten et al. [18], the classical diagnostic techniques (including histopathology) have limitations in detecting minimal residual disease. The most commonly used nucleic acid markers are mutations in tumor DNA (tumor-specific markers) or differentially expressed RNA (tissue-specific markers). There are contradictory findings in the literature on molecular markers of minimal disease, and these are related to the difficulties in finding small numbers of cancerous cells among the much larger number of normal cells. (See Ref. [18] for a discussion of the technical issues that need to be addressed as they relate specifically to molecular assays diagnosing minimal residual head and neck cancer).

Mutated p53 In a subsequent paper, Van Houten et al. [19] further examine the problem of minimal residual cancer in 50 patients with head and neck tumors, who had undergone radical dissection. They used mutated p53 as a marker. Histopathologically tumor-free surgical margins were assessed quantitatively for the presence of mutated p53 using plaque assay and immunohistopathology. Mutated p53 was present in 19 of the 30 patients. By immunohistopathology, small tumor foci were found in 2 of these 19 patients. In 7 of the 19 patients, tumor-specific mutated p53 was found in unresected dysplastic precursor lesions. They demonstrated that using mutated p53 as a marker lead to a sizeable number of false positive results. They conclude that molecular assessment of surgical margins using p53 mutations can help identify patients with head and neck tumors at high risk for cancer recurrence. However, they consider tumor RNA to be a more specific biomolecule than tumor DNA.

Classification and Grading Squamous cell carcinomas of the head and neck are classified as: well differentiated, moderately well differentiated and poorly differentiated, with prognosis affected by degree of differentiation [1]. There are, however, some difficulties and controversies classification, and the degree to which this predicts prognosis [20]. in

188

Staging of head and neck tumors is according to the TNM system. About one-third of patients will have localized disease Stage I or II, without lymph node involvement or metastases. The staging procedure has included chest X-ray and bone scan if lymph nodes are involved. Definitive staging has required endoscopic examination under anesthesia, and may include esophagoscopy, bronchoscopy and laryngoscopy. If there is lymph node involvement without an identifiable primary, lymph node excision should be performed, and if squamous cell carcinoma is found, pan-endoscopy should then be carried out [1].

Treatment and Prognosis

Prognosis depends on stage at diagnosis [15]. Localized disease (Stage I or II):
Surgery Preferred for small lesions in the oral cavity to avoid long-term complications of radiation. Radiation therapy RT is preferred for laryngeal cancerto preserve voice function.

Both therapeutic modalities are frequently curative and can produce similar cure rates. Selection of treatment must be individualized considering cosmetic, functional, quality of life, capacity for salvage therapy, patient reliability, inter alia [1]. Locally or regionally advanced disease This is defined as a large primary tumor possibly with regional lymph node involvement. Locally or regionally advanced disease can be treated with curative intent, but requires multi-modal therapy (surgery, RT and chemotherapy) [1]. Measures to prevent a second primary As noted earlier in this chapter, patients who have been successfully treated for head and neck tumors often develop a second primary tumor in the head and neck region. This is frequently preceded by leukoplakia, and represents a major cause of treatment failure. Preventive measures include tobacco and alcohol abstinence, surveillance endoscopy and secondary chemoprevention to suppress or reverse the malignant process [21].

189

Chemoprevention Anderson et al. [21] note: Classic antioxidant micronutrients such as retinoids, carotenoids and certain other agents have been effective in nonrandomized and randomized clinical trials, but treatment is uncertain and recurrences common. These facts, coupled with recent harmful effects of betacarotene in two clinical trials, stress the need for additional basic science, translational and clinical research. Chemoprevention is a promising new technology, but is not currently standard therapy for the secondary prevention of {head and neck} tumors (p. 106). Retinoids have a documented effect upon head and neck malignancies, but also have major side effects that render them unacceptable [22]. These include acute mucocutaneous toxicity as well as gastrointestinal, neurological and musculoskeletal effects and they are teratogenic [23]. The retinoids are analogues of Vitamin A that can modulate the growth and differentiation of normal, pre-cancerous and cancerous epithelial cells in vitro, and can suppress carcinogenesis in vivo in human epithelial tissues [16]. Cyclooxygenase inhibitors and epidermal growth factor receptor inhibitors are possible agents under investigation for chemoprevention [22].

Rehabilitation after treatment for head and neck cancer Miller and Sessions [24] stress the critical importance of counselling and the need for a multi-disciplinary team involved in rehabilitation, the details of which are outlined in their review.

11.2 Anatomical and Functional Imaging

11.2.1 FDG-PET and CT PET evaluation of the head and neck region, particularly after surgery, has become an important diagnostic modality. Head and neck cancers as small as 4 mm have been detected using FDG-PET. However, after radiation therapy there are frequent false positive results due to inflammation, and false negatives related to possible metabolic stunning. Therefore, it is recommend to wait 40 days to 4 months or more after therapy to apply FDG-PET [25].

11.2.2 MRI MRI and CT have been used for staging and treatment planning [2]. By providing multi-planar imaging, MRI is advantageous compared to CT, and has been reported to better detect tumor extensions [20].

190

Comparison between MRI and CT A comparison from Ref. [26] of MRI with Gadolinium-DTPA4 and CT in 1260 patients with lesions of the head and neck revealed different enhancement patterns that helped distinguish malignant from benign lesions (cysts, inflammatory). With the exception of lesions involving small bony erosion or inflammation of the salivary glands, MR with Gd-DTPA was diagnostically advantageous.

MRI for N0 Neck (no palpable lymph nodes) Yucel et al. [27] applied MRI in 18 patients with squamous cell carcinoma of the head and neck and with no palpable lymph nodes (socalled N0 neck). This was compared with histopathology. MRI suggested node involvement in 5 patients, 2 of whom had central necrosis, and in another patient the nodes were grouped. Histopathology revealed positive nodes in 7 patients, 4 of whom were detected by MRI. There was 1 false positive from MRI. The authors conclude that MR can reveal lymph node involvement in patients without clinical evidence of such. They note however, that conventional MR techniques are not always sufficient for decisionmaking on surgery in cases of N0 neck. Ishikawa and Anzai [28] consider that improved accuracy in imaging lymph nodes of the head and neck will require new tissue-specific MR contrast agents and functional imaging to evaluate biological activity.

Dynamic Contrast Enhanced MRI Dynamic MRI is helpful for planning and assessing RT in patients with head and neck tumors [20]. This technique takes advantage of different temporal enhancement characteristics of tumors compared to adjacent non-malignant tissue, which may contain edema and inflammation. These differences are related to vascularity, capillary permeability, renal clearance and extracellular fluid composition and volume [29]. In a study of 22 patients with head and neck cancers, dynamic gradient-echo imaging showed superior or similar tumor margin delineation compared to the more commonly used spin-echo MRI technique [29] in all but two cases, enabling clinicians to more confidently interpret findings, as well as having fewer technical limitations.

DTPA is the acronym for the chelating agent diethylenetriaminepentaacetic acid

191

Dynamic contrast-enhanced MRI has also been shown to be helpful and feasible for detecting tumor involvement in cervical lymph node among patients with head and neck cancers [30]. Most cancers of the head and neck show early enhancement and early washout of contrast media on dynamic MRI. Pharmocokinetic study of dynamic MRI can give insight into the permeability of gadolinium in the tumor that may reflect oxygenation and amount of medication delivered to the tumor. This may be helpful in predicting response to RT and chemotherapy [31]. Tissue Perfusion Imaging with MRI Exploiting the fact that tumor blood flow, tissue perfusion and oxygen supply are determinants of responsiveness to RT, a study [32] of quantitative tissue perfusion among patients with head and neck cancers before and after RT was performed. This technique provided visualization of tissue perfusion with a resolution < 3 mm. In four of the five examined patients, perfusion decreased after RT, but in one patient perfusion increased. The authors consider that tissue perfusion MR imaging may be helpful for treatment stratification, in particular for anti-angiogenic or other vasomodulatory agents. 11.2.3 In vivo MRS for primary detection and assessing response to therapy The potential role of MRS in head and neck tumors has been highlighted by El-Sayed et al. [20], since the heterogeneity of neoplastic processes in the upper aerodigestive tract {and} the difficulty in the interpretation of histopathologic findings {has prompted} the search for non-invasive diagnostic procedures, and the pursuit of new prognostic and treatment selection factors (p. 767). Proton MRS Mukherji et al. [33] performed in vivo proton MRS to compare seven patients with squamous cell carcinoma of the upper aerodigestive tract (oral cavity, oropharynx, nasopharynx, six with primary and one with recurrent disease), and seven healthy volunteers. These authors5 found a high choline to creatine ratio in all the patients on 1D proton MRS; while in six of the seven healthy participants there were no detectable choline resonances.
5 Mukherji et al. [33] used 1.5T, PRESS, single-voxel, TE=136 ms, estimated peak areas as the product of peak height and line-width at half-maximum height, ratios calculated. Since a head coil was used, only tumors of the oral cavity, oropharynx and nasopharynx were assessed, not those from the larynx or hypolarynx.

192

Star-Lack et al. [34] used proton MRS6 to assess lymph node metastases in 14 patients with head and neck cancer, and compared these spectra to those from neck muscle tissue of six healthy volunteers. They also examined tissue oxygenation with a polarographic oxygen electrode. These authors found that choline to creatine ratios were significantly higher in the nodes (2.9 1.6) than in the normal muscle (0.55 0.21) (p = 0.0006), and lactate was significantly higher in the cancerous nodes than in muscle (p = 0.01). There was a negative correlation between tissue oxygenation and lactate. The authors conclude that proton MRS may be helpful in identifying nodal involvement in head and neck cancer and assessing oxygenation, and this might be applied in staging and monitoring therapy.
31

P MRS

There have been two early studies of 31P MRS in head and neck patients. One of these (Ref. [35]) included 30 patients with squamous cell tumors of the head and neck and showed Pi and PME, which are not usually seen in normal muscle, elevated PME to -ATP ratios > 1 and PME > PDE7. A change in the 31P spectrum was seen during and after RT; these changes were associated with alterations in the metabolic activity of the tumor, and/or changes in lipoprotein metabolism. The authors suggested that MRI and MRS might be helpful in RT planning and monitoring patients with head and neck tumor before and after therapy. In the same year, another investigation [36] was published of 15 patients with superficial masses (sarcoma, carcinoma, lymphoma, adenoma and tuberculosis), who were studied using 31P MRS. Tumor growth was associated with increased PME concentrations. During RT of a squamous cell carcinoma, there was a substantial fall in Pi and a subsequent rise in PDE. 11.2.4 In Vitro MRS studies Using 2D correlated proton MRS in vitro, in the above-cited study, Mukherji et al. [33] identified a number of amino acid resonances (alanine, glutathione, histidine, isoleucine, valine, and the shared cross peak for lysine/polyamine) in most or all of the malignant specimens that were only rarely detected in normal tissue. More recently, El-Sayed et al. [20] examined a total of 135 tumor and adjacent normal tissue specimens from untreated patients with head and neck cancers from various sites. They found that taurine,
6 Star-Lack et al. [34] used 1.5T, PRESS, single-voxel, TR/TE=2000/144 ms, peak areas calculated, 3.1-3.3 ppm for choline and 2.9 to 3.1 ppm for creatine. Lactate and lipids peak areas were calculated by integrating the respective coupled and uncoupled spectra from 1.5 to 1.1 ppm. 7 PDE is the acronym for phosphodiester.

193

choline, glutamate, lactate and lipid were the most important metabolites for differentiating malignant from normal tissue. Using multivariate analytical techniques, overall accuracy was 92.6% (training set 97.3%, and test set 87.3%). Taurine was seen in most of the malignant spectra, and seemed to be associated with higher proliferative activity. The choline to creatine ratio was generally higher in the malignant spectra. Most of the malignant spectra resembled each other: reflective of the common neoplastic process (p. 769).

References
[1] E.E. Vokes, Head and neck cancer, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 559 562. [2] L. Mucci, H-O Adami, in: H-O. Adami, Oral and pharyngeal cancer, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, Oxford, 2002, p.115136. [3] Z.P. Pavelic, J.L. Gluckman, The role of p53 tumor suppressor gene in human head and neck tumorigenesis, Acta Oto-Laryngol. 527 (Suppl.), 21-24 (1997). [4] D. Sidransky, Molecular biology of head and neck tumors, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 718-796. [5] K.J. Rothman, Epidemiology of head and neck cancer, Laryngoscope 88, 435-438 (1978). [6] K. Rothman, A. Keller, The effect of joint exposure to alcohol and tobacco on risk of cancer of the mouth and pharynx, J. Chronic Dis. 25, 711-716 (1972). [7] S. Sriamporn, V. Vatanasapt, P. Pisani, S. Yongchaiyudha, V. Rungpitarangsri, Environmental risk factors for nasopharyngeal carcinoma: a case-control study in northeastern Thailand, Cancer Epid. Biomarker, Prev. 1, 345-348 (1992). [8] I. Kawachi, N. Pearce, J. Fraser, A New Zealand Cancer Registry-based study of cancer in wood workers, Cancer 64, 2609-2613 (1989). [9] T.P. Ng, A case-referent study of cancer of the nasal cavity and sinuses in Hong Kong, Int. J. Epidemiol. 15, 171-175 (1986). [10] J.H. Wills, Nasal carcinoma in woodworkers: a review, J. Occup. Med. 24, 526-530 (1982). [11] W. Zheng, J.K. McLaughlin, H.Y.T. Gao, R.N. Gao, W.J. Blot, Occupational risks for nasopharyngeal cancer in Shanghai, J. Occup. Med. 34, 1004-1007 (1992). [12] Y.T. Wu, H.L. Luo, D.R. Johnson, Effect of nickel sulfate on cellular proliferation and Epstein-Barr virus antigen expression in lymphoblastoid cell lines, Cancer Letters 32, 171-179 (1986). [13] K.S. Kasprzak, F.W. Sunderman, K. Salnikow, Nickel carcinogenesis, Mutation Res. 533, 67-97 (2003). [14] F.W. Sunderman, Nasal toxicity, carcinogenicity, and olfactory uptake of metals, Ann. Clin. Lab. Sci. 31, 3-24 (2001).

194

[15] I.B. Tan, J. L. Roodenburg, M.P. Copper, J.W. Coebergh, I. van der Waal, Early diagnosis and prevention of malignant tumors in the head and neck region, Nederlands Tijdschr. Geneesk. 145, 567-572 (2001). [16] S.P. Schantz, L.B. Harrison, A.A. Forastiere, Tumors of the nasal cavity and paranasal sinuses, nasopharynx, oral cavity and oropharynx, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p.797-860. [17] C.R. Smart, Screening for cancer of the aerodigestive tract, Cancer 72 (Suppl.) 1061 (1993). [18] V.M.M. van Houten, M.P. Tabor, M.W.M. van den Brekel, et al. Molecular assays for the diagnosis of minimal residual head-and-neck cancer: methods, reliability, pitfalls, and solutions, Clin. Cancer Res. 6, 3803-3816 (2000). [19] V.M.M. van Houten, M.P. Tabor, M.W.M. van den Brekel, et al. Mutated p53 as a molecular marker for the diagnosis of head and neck cancer, J. Pathol. 198, 476-486 (2002). [20] S. El-Sayed, T. Bezabeh, O. Odlum et al. An ex vivo study exploring the diagnostic potential of 1H magnetic resonance spectroscopy in squamous cell carcinoma of the head and neck region, Head & Neck 24, 766-772 (2002). [21] W.F. Anderson, E. Hawk, C.D. Berg, Secondary chemoprevention of upper aerodigestive tract tumors, Semin. Oncol. 28, 106-120 (2001). [22]. J. Sudbo, Chemoprevention of oral cancer, Tidskr. Norske. Laegef. 123, 1518-1521 (2003). [23] A.R. Brecher, S.J. Orlow, Oral retinoids therapy for dermatologic conditions in children and adolescents, J. Am. Acad. Dermatol. 49, 171-182 (2003). [24] S.D. Miller, R.B. Sessions, Rehabilitation after treatment for head and neck cancer, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 907-916. [25] M. G. Pomper, Functional and metabolic imaging, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 679-689. [26] T. Vogl, S. Dresel, M. Juergen, J. Assal, J. Lissner, MR imaging with Gd-DTPA in lesions of the head and neck, J. Otolaryngol. 22, 220-230 (1993). [27] T. Yucel, I. Saatci, L. Sennaroglu, S. Cekirge, U. Aydingoz, S. Kaya, MR imaging in squamous cell carcinoma of the head and neck with no palpable nodes, Acta Radiol. 38, 810-814 (1997). [28] M. Ishikawa, Y. Anzai, MR imaging of lymph nodes in the head and neck, Magn. Reson. Imaging Clin. N. Am. 10, 527-542 (2002). [29] E.J. Escott, V.M. Rao, W.D. Ko, J.E. Guitierrez, Comparison of dynamic contrast-enhanced gradient-echo and spin-echo sequences in MR of head and neck neoplasms, Am. J. Neuroradiol. 18, 1411-1419 (1997). [30] S.M. Noworolski, N.J. Fischbein, M.J. Kaplan, et al., Challenges in dynamic contrastenhanced MR imaging of cervical lymph nodes to detect metastatic disease, J. Magn. Reson. Imaging 17, 455-462 (2003). [31] Y. Baba, Y. Yamashita, M. Onomichi, R. Murakami, M. Takahashi, Dynamic magnetic resonance imaging of head and neck lesions, Top. Magn. Reson. Imaging 19, 125-129 (1999). [32] P. Schmidtt, M. Kotas, A. Tobermann, A. Haase, M. Flentje, Quantitative tissue perfusion measurements in head and neck carcinoma patients before and during radiation therapy with a non-invasive MR imaging spin-labeling technique, Radiother. Oncol. 67, 27-34 (2003).

195

[33] S.K. Mukherji, S. Schiro, M. Castillo M, et al,, Proton MR spectroscopy of squamous cell carcinoma of the extracranial head and neck: in vitro and in vivo studies, Am. J. Neuroradiol) 18, 1057-1072 (1997). [34] J.M. Star-Lack, E. Adalsteinsson, M.F. Adam, et al., In vivo 1H MR spectroscopy of human head and neck lymph node metastasis and comparison with oxygen tension measurements, Am. J. Neuroradiol. 21, 183-193 (2000). [35] W.G. McKenna, R.E. Lenkinski, R.A. Hendrix, K.E.Vogele, P. Block, The use of magnetic resonance imaging and spectroscopy in the assessment of patients with head and neck and other superficial human malignancies, Cancer 64, 2069-2075 (1989). [36] T. Vogel, F. Peer, H. Schedel, et al., 31P-spectroscopy of head and neck tumorssurface coil technique, Magn. Reson. Imaging 7, 425-435 (1989).

197

Chapter 12

Non-Hodgkins Lymphoma
_______________________________________________________________________________

Non-Hodgkins lymphomas comprise a heterogeneous group of lymphoid malignancies [1]. As succinctly stated by Melbye and Trichopoulos [2]: Non-Hodgkins lymphoma (NHL) is unique among human malignancies for several reasons. First, it is distinguished as an entity by exclusion from a much less common group of lymphomas called Hodgkins lymphoma. Secondand most importantthe incidence of NHL is rising more rapidly than that of virtually any other human cancer in many parts of the world. Indeed, this increase has been considered almost epidemic and its causes remain enigmatic, though it suggests exposure to new causal agents that might be avoidable. Finally, no other cancer entity comprises such a heterogeneous group of malignancies, as does NHL (p. 535).

12.1 Overview of epidemiological & clinical aspects

Incidence and prevalence/morbidity and mortality Non-Hodgkins lymphomas are the seventh most commonly diagnosed cancer, accounting for about 2.5% of all malignancies. Gender and racial differences in occurrence depend upon the type, but overall incidence is higher among men and increases with age [2]. Time trends reveal a sharp rise in incidence of NHL in western counties. On the average, there has been a 3-4% increase per year, especially among young and middle-aged persons. Rising incidence has also been seen in Japan, Singapore, India, Brazil and Puerto Rico. In the U.S. the incidence of NHL has been increasing since the 1950s at the rate of about 4%. The reason for this rise is not known. In the year 2000, there were nearly 60 000 new cases of NHL diagnosed in the

198

U.S. These reported trends cannot be explained by changes in classification [2, 3]. There are marked geographic differences in the rates of occurrence of the various sub-types. T-cell lymphomas are more common in Asia; follicular lymphoma is more common in western countries. The angiocentric nasal T / natural killer (NK) cell lymphoma is most frequent in Southern Asia and in parts of Latin America. The NHL subtype related to infection with HTLV-11 is most often seen in southern Japan and in the Caribbean [3]. In the U.S. there has been a rising incidence in all geographic regions, with the largest increase in the San Francisco area. High rates of gastric lymphoma are found in Northern Italy. The incidence of follicular lymphoma is lower among Asian immigrants to the U.S. compared to subsequent generations. A possible environmental influence is therefore postulated [4]. Extra-nodal NHL comprises about 25-30% of all NHL, with brain and skin showing the highest proportional increase. This may be related to human immunodeficiency virus (HIV) infection as well as to improvements in diagnosis. The B-cell phenotype is predominant in most parts of the world [2]. Etiology/risk factors Genetic and Molecular Epidemiology There is an observed familial aggregation of NHL. Having a near relative with NHL or other hematolymphoproliferative cancer confers an estimated 2.5 to 4-fold increased risk [2]. NHL is associated with a number of genetic abnormalities. Many of these are balanced chromosomal translocations involving antigen receptor genes, immunoglobulin genes for B-cell lymphomas, and T-cell antigen receptor genes for T-cell lymphomas. B-cells are vulnerable to acquiring mutations during maturation in germinal centers [3]. The t (14; 18) translocation is seen in about 30% of patients with diffuse large B-cell lymphomas. This leads to over-expression of the bcl-2 gene on chromosome 18; this protein suppresses apoptosis. Patients with over-expression of bcl-2 have a higher relapse rate than those who only have translocation. The t (14; 18) translocation is also associated with follicular lymphomas. A number of other cytogenetic abnormalities and associated oncogenes have been related to specific types of NHL [3]. Increased risk for NHL has been demonstrated for inherited immunodeficiency disorders:

HTLV 1 is the acronym for Human T-cell leukemia/lymphoma virus l

199

Wiscott-Aldrich syndrome This is an X-linked recessive syndrome, caused by mutations in the WASP gene, which plays an important role in T-cell and platelet function. It is associated with low IgM, anergy, thrombocytopenia, eczema and repeated infections. Ataxia-telangiectasia (AT) AT is an autosomal recessive disorder, characterized by a deficit in cerebellar function, nystagmus, genetic deficiency in monitoring DNA repair and coordination of DNA synthesis with cell division. Individuals with AT are highly susceptible to radiation-induced chromosomal damage, have decreased IgE and IgA and reduced T-cell pool. Other abnormalities include chronic pulmonary disease and scleral telangiectasia. The heterozygous state is associated with increased radiosensitivity and predisposition to cancer. Chdiak-Higashi syndrome This is an autosomal recessive syndrome, characterized by fusion of cytoplasmic granules, defective degranulation of neutrophil lysosomes, and associated with NK hyporesponsiveness. Common variable immunodeficiency disorder This disorder is characterized by deficient production of all immunoglobulins with phenotypically immature B-lymphocytes. The susceptibility gene is in the major histocompatibility complex class III region. This disorder may progress from IgA deficiency.

Infections
Epstein-Barr virus Burkitts lymphoma is highest among children in regions of Africa that are endemic for malaria. It is suggested that Burkitts lymphoma in Africa may be due to increased B-cell proliferation from early EBV infection interacting with the mitogenic effects of malaria. Over 95% of cases of Burkitts lymphoma in Africa show EBV DNA. In developing countries outside Africa, 50-90% of cases of Burkitts lymphoma are EBV positive. However, in temperate areas of South America, this association between Burkitts lymphoma and EBV is weaker. Burkitts lymphoma is also linked to alterations of the c-Myc gene. However, the role of EBV in c-Myc gene transformation is not known [2]. Among immunocompetent patients, less than 5% of other NHL of Bcell origin contain EBV DNA. However, in patients with primary immunodeficiency syndromes, EBV DNA is nearly always present. Among HIV-infected persons, about 50% show monoclonal EBV, and among those with CNS involvement, nearly all contain monoclonal EBV [2]. Among immunosuppressed patients, EBV is associated with aggressive NHL. EBV is also linked to extranodal nasal T/NK cell lymphomas in Asia and South America [3].

200

Human immunodeficiency virus HIV infection is associated with aggressive, B-cell NHL. This may be related to over-expression of interleukin 6 by infected macrophages [3] (see later subsection on immune modulation).
Human T-cell leukemia/lymphoma virus l There is strong evidence that HTLV-1 is a causative agent of adult T-cell leukemia/lymphoma, which is common in Japan and the Caribbean [2, 3]. Human herpesvirus 8 Associated with Kaposis sarcoma, seen in primary effusion lymphoma, a rare type of NHL seen only among persons who are HIV-positive [2]. Chronic infection with hepatitis C virus Associated with lymphoplasmacytic lymphoma [3]. Infection with hepatitis C virus is strongly associated with essential mixed cryoglobulinemia, which is, in turn, associated with low-grade NHL. The hepatitis C virus causes chronic antigenic stimulation, which can lead to clonal expansion and NHL [4]. Helicobacter Pylori (H. Pylori) Primary gastric low-grade lymphoma of the mucosa-associated lymphoid tissue (MALT) is induced by H. pylori infection. The bacterium stimulates a strong immune response and it is the chronic antigenic stimulation, which leads to this type of NHL. Eradication therapy has been found in individual patients to be associated with partial or complete remission of MALT lymphoma [3, 5]. Borrelia burgdorferi Can cause lymphoproliferation (pseudo-lymphoma), which responds to antibiotics. Cutaneous B-cell lymphoma has shown an association in Europe (but not in North America) with Borrelia burgdorferi infection. This may be due to differences in the agent between the two continents [2]. Chronic infectious disease There is fairly consistent evidence that chronic infectious diseases such as tuberculosis, malaria, herpes zoster, chronic otitis, bronchitis and pyelonephritis are related to increased risk of NHL. This could be plausibly explained as a result of chronic antigenic stimulation of B-cells by chronic infection [2]. By logic, the authors from Ref. [2] suggest: our ability today to treat disease that previously might have been lethal, such as tuberculosis and other chronic infections, may have led to an increasing pool of subjects at higher risk of NHL development. If this is true, such a situation might very well explain the observed worldwide increasing incidence of NHL (p.548).

201

Occupations and work-related exposures There are difficulties in etiologic studies on occupation in relationship to B-lymphocyte malignancies, due to inconsistencies in the classification system of lymphohematopoietic cancers [6].
Agricultural work - exposure to pesticides The most extensive investigations have been of farmers. Exposure to pesticides has increased steadily since the 1940s, and it has been suggested that chromosomal rearrangements associated with pesticide application could be related to increased risk of NHL [2]. An investigation [7] of death certificates from 26 of the states of the U.S. from 1984 to 1993 reveals cause-specific proportionate mortality ratios among white male livestock farmers to be elevated for NHL, as well as multiple myeloma, acute and chronic lymphoid leukemia and Parkinsons disease. These disease trends suggest that livestock farmers may be exposed to more carcinogens or agricultural chemicals compared to crop farmers. Among women in England and Wales, a review of the cancer registries from 1971 to 1990 [8] reveals a proportional registration ratio of 164 (95% CI = 126 - 211) for NHL among agricultural workers. A case-control study [9] from Kansas and Nebraska, U.S. of 555 incident cases of NHL, 56 of chronic lymphocytic leukemia (CLL) and 2 380 population-based referents reveals an increased risk in men for NHL and CLL in agricultural, forestry and logging occupations (OR = 1.6, 95% CI = 1.2 - 2.1). The OR for those involved in producing crops was 1.9 (95% CI = 1.4 - 2.6). These associations strengthened with duration of employment. However, in a study [10] of an employed Swedish cohort followed for 19 years, with 7 610 incident cases of NHL there was no increased risk observed among agricultural occupations, unlike the findings of other studies.
t (14; 18) translocations

As mentioned earlier, t (14; 18) translocations are a common somatic mutation in NHL, associated with bcl-2 activation and inhibition of apoptosis. Schroeder et al. [11] examined archival biopsies of 182 cases of NHL in Iowa and Minnesota for t (14; 18), finding that 68 (37%) were positive. They report that the t (14; 18) translocation positive NHL was associated with farming (OR = 1.4, 95% CI = 0.9 2.3), with dieldrin (OR = 3.7, 95% CI = 1.9 7.0), toxaphene (OR = 3.0, 95% CI = 1.5 6.1), lindane (OR = 2.3, 95% CI = 1.3 3.9), atrazine (OR = 1.7, 95% CI = 1.0 2.8), and fungicides (OR = 1.8, 95% CI = 0.9 3.6). This was in sharp contrast to null or negative findings for these self-reported exposures and t (14; 18)-negative NHL. The authors conclude: causal relations between agricultural exposures and t (14; 18)positive non-Hodgkins lymphoma are plausible, but associations should be confirmed in a larger study. Results suggest that non-Hodgkins lymphoma classification based on the t (14; 18) translocation is of value in etiologic research (p. 701). Roulland et al. [12] examined a representative sample of 56 persons with occupational exposure to pesticides from farming, noting that t (14; 18) translocations that deregulate bcl-2 expression and inhibit apoptosis, are

202

frequently found in follicular lymphoma, and that bcl-2 IGH translocation is an early step in neoplasia. Their results suggest that occupational exposure to pesticides could increase bcl-2 IGH prevalence. This could be used to assess acquired genetic instability in relation to genotoxic exposure. Industrial and other occupations Taking into consideration methodological limitations, Bukowski et al. [6] observe in their review that there is a pattern of B-cell cancer elevations in the rubber and general chemical industries. In the previously cited Swedish cohort study by Cano and Pollan [10] a relative risk > 1.2 was found among men employed as accountants and auditors, secretaries and typists, auctioneers, railroad and transport workers, telecommunications traffic officers, telegraph and radio operators, photographic-laboratory workers and other production workers. Among women, the excess risk was for metal platers and coaters, truck and conveyor operators, and store and warehouse workers. The authors attribute the risk in telecommunication and transport workers as possibly due to exposure to electromagnetic radiation. There has also been an increased risk reported among those in metalworking and motor vehicle industries, as well as welders and solderers [9]. In a case-control study from Italy [13] with 2 737 of 3 357 incident cases included and 1 779 of 2 391 eligible controls, NHL risk in men was increased among cooks, waiters, bartenders, building caretakers and cleaners. The authors note that an increased risk of NHL among cooks, waiters and bartenders had not been reported previously. In a study from Canada [14], 1 469 newly diagnosed patients with histopathologically confirmed NHL and 5 073 population-based referents were queried by questionnaire about occupation and exposures to chemicals, inter alia. An elevated risk of NHL was found among men exposed to benzidine (adjusted OR = 1.9 (95% CI = 1.1 3.4)), mineral, cutting or lubricating oil (adjusted OR = 1.3 (95% CI = 1.0 1.5)), pesticides (adjusted OR = 1.3 (95% CI = 1.0 1.6)) and herbicides (adjusted OR = 1.3 (95% CI = 1.0 1.6)). Women with high risk had been exposed to pesticides or wood dust. There were significant trends (p < 0.05) for duration of exposure to benzidine and herbicides among men, and for wood dust among women. The authors [14] conclude: these findings suggest that occupational exposure to specific chemicals plays an important role in the development of NHL in Canada (p.
69).

Medical conditions and medical-related exposures

Immune modulation There is a strong relationship between immune modulation and NHL. As noted earlier, elevated risk is found among patients with primary and acquired immunodeficiency. This is also the case for patients receiving immunosuppressive therapy. However, the level of antigenic stimulation is also important, and these two factors in organ transplantation are difficult to disentangle. In the first year after organ transplantation there is a 20-fold increased risk of NHL; this falls to 5-10 fold subsequently [2].

203

Among HIV-infected persons, the relative risk of NHL is about 100 fold increased. In patients with AIDS, the risk of NHL is closely (inversely) related to the CD4 count2. With treatment using highly active antiretroviral therapy for patients who are HIV-positive, immunocompetence is markedly improved, and NHL incidence is lowered. It has been suggested that the level of immunosuppression could affect immunological control over lymphoproliferative growth. With HIV-positive status, non-lethal genetic errors could thereby accumulate, together with chronic antigenic stimulation of lymphocytes that are responding to HIV-related infections [2]. Autoimmune disease A number of autoimmune diseases: rheumatoid arthritis, systemic lupus erythematosis, Sjgrens syndrome, celiac disease, and possibly dermatitis herpetiformis have all been associated with increased risk of NHL. Although immunosuppressive therapy contributes to this risk, there is also evidence of a direct relationship with the presence of these diseases. The increased risk associated with rheumatoid arthritis and Sjgrens syndrome also is seen among patients not treated with immunosuppressives. High inflammatory activity is associated with particularly elevated risk [2]. In a review by Varoczy et al. [15] of the clinical records of 421 patients with NHL in Debrecen, Hungary, 32 (7.6%) were found to have a diagnosed autoimmune disease, with Sjgrens syndrome being the most common. Medications Increased risk of NHL has been reported in relation to the following medications [2], inter alia: Dilantin (inhibited delayed hypersensitivity responses, can produce lymphadenopathy that can progress to lymphoma) Penicillin and other antibiotic used for > 2 months Digitalis Tranquilizers (mebrobamate) Tricyclic antidepressants. Other medical conditions or treatment The empirical data, though not entirely consistent, suggest that adult-onset diabetes mellitus is associated with an increased risk of NHL [2]. In the above-cited study of Varoczy et al. [15] insulin-dependent diabetes mellitus was a co-morbid finding among patients with Hodgkins lymphoma, but not among those with NHL. However, van Spronsen et al. [16] reported that among a series of 904 patients with NHL aged 60 and over, diabetes mellitus was one of the most frequent co-morbidities. Prior exposure to chemotherapy or RT is associated with increased risk of NHL [3].

2 CD denotes cluster of differentiation, and is a classification of human lymphocyte differentiation antigens. The CD4 surface antigen is the primary receptor for HIV and is involved in T-cell activation and function.

204

Other possible risk factors

Ultra-violet light It has been suggested that in order to explain the dramatic rise in incidence of NHL, exposures shared by a large number of populations across the world would need to be considered. One of these is increasing exposure to ultraviolet light due to depletion of the ozone layer, and due, as well, to individual behavior. Animal studies reveal an elevated occurrence of lymphomas after UV light exposure. A highly significant association has been found between hprt mutant frequency and sunlight exposure in the immediate period. British migrants to Australia show an increased risk of NHL as well as malignant melanoma. There is also an association between skin cancer (as a surrogate measure of exposure to UV light) and NHL. However, there are also null studies with respect to this hypothesis [2].

Perhaps a fruitful line of future investigation would be to consider combined exposures [17]. Then, for example, farmers as a likely highrisk occupational group for NHL would be evaluated vis--vis their exposure not only to pesticides, but also to high levels of UV light, inter alia. Melbye and Trichopoulos [2] point out: so far, most studies addressing risk factors for NHL have been limited by their small size and their focus on NHL as a single entity. Recent evaluations of epidemiologic patterns of NHL according to histologic subtype have documented clear differences among these sub-types future research on possible risk factors for NHL needs to include larger studies permitting analyses by histologic subtype (p. 550). Clinical Presentation and Approach The most frequent early manifestations of NHL are painless lymphadenopathy, fever, night sweats and weight loss. Also seen are anorexia, fatigue, dermatologic manifestations including pruritus, bleeding skin and/or erythematous patches and bone pain. However, early NHL is often completely asymptomatic and diagnosed during routine medical procedures [2]. General approach to the patient with suspected NHL Regardless of the type of lymphoid malignancy, the initial evaluation of the patient should include performance of a careful history and physical examination. These will help confirm the diagnosis, identify those manifestations of the disease that might require prompt attention, and aid in the selection of further studies to optimally characterize the patients status to allow the best choice of therapy. It is difficult to overemphasize the importance of a carefully done history and physical examination. They might provide observations that lead to reconsidering the diagnosis, provide hints at etiology, clarify the stage,

205

and allow the physician to establish rapport with the patient that will make it possible to develop and carry out a therapeutic plan (p. 718) [3]. In patients with NHL the evaluation includes: complete blood count, sedimentation rate, laboratory assessment of liver function, as well as serum calcium, uric acid, lactate dehydrogenase, 2microglobulin and protein electrophoresis, CT of the chest, abdomen and pelvis, bone marrow biopsy, with additional examinations based upon sub-type [3]. Histopathologic diagnosis is usually made from lymph node biopsy. For sub-grouping, immunohistochemistry, flow cytometry, cytogenetics, or molecular genetic studies are usually needed [2]. Diffuse, large B-cell lymphoma is the most common type of NHL. It comprises the majority of patients in clinical trials of intermediate or aggressive lymphomas. It can present as primarily lymph node disease or at extranodal sites. Histopathologic diagnosis is very important, because, e.g. with pancreatic involvement, differential diagnosis would include primary pancreatic carcinoma which has a much worse prognosis. Primary diffuse large B-cell lymphoma of the brain is being diagnosed more and more frequently [3] (see Chapter 8). Differential Diagnosis Cell-surface phenotyping helps to distinguish e.g. benign follicular hyperplasia from NHL, both of which appear similar on light microscopy. Whereas benign hyperplasia is polyclonal, follicular lymphoma will show the same immunoglobulin light chain isotype [3]. Conditions often confused with NHL by clinicians as well as by pathologists, include: Reactive, atypical lymphoid hyperplasia, which can be due to a reaction to dilantin or carbamezapine, Lymphadenopathy associated with rheumatoid arthritis, lupus erythematosis, viral infections including EBV, cytomegalovirus, and bacterial infections such as cat-scratch disease,

Castlemans disease Lymphadenopathy accompanied hypergammaglobulinemia and, glucocorticoids,

by if

anemia and polyclonal systemic, treated with

Sinus histiocytosis with massive lymphadenopathy (RosaiDorfmans disease), which is self-limited.

206

Depending upon the site of involvement, primary carcinoma, sarcoma or metastatic carcinoma can be in the differential diagnosis of NHL. It should also be recalled that many of these conditions considered in the differential diagnosis are also associated with increased risk of NHL. Classification and Grading There have been many classification systems used, with substantial difficulties. The WHO adopted in 2000 the Revised EuropeanAmerican Classification of Lymphoid Neoplasms. The WHO classification takes into account morphologic, clinical, immunological and genetic information. This classification system attempts to divide NHL into clinical/pathological entities that are relevant for clinical decision-making, especially with regard to therapy [3]. This includes Bcell malignancies (85%), T/NK-cell neoplasms and Hodgkins lymphoma. There are several specific entities frequently associated with certain karyotypic changes.

Mature B-cell malignancies These include follicular lymphomas, extranodal marginal zone B-cell lymphoma of MALT type, diffuse large B-cell lymphomas and Burkitts lymphoma, inter alia. The two most common histological types are follicular and diffuse large B-cell lymphomas [1]. Mature T and NK-Cell malignancies Included here are hepatosplenic T-cell lymphomas, the extranodal NK/T-cell lymphoma of the nasal type, anaplastic large cell lymphoma and primary cutaneous type, inter alia [2].
The stage of differentiation of a malignant lymphoma does not describe its natural history. Burkitts lymphoma, e.g. is phenotypically a mature follicle center IgM-bearing B-cell lymphoma, but is highly aggressive. Cell-surface phenotyping is useful mainly for diagnostic purposes [3]. Also used, at least in part, for NHL is the Ann Arbor Staging System for Hodgkins Disease [3]:
Stage I = Involvement of a single lymph node region or lymphoid structure (e.g. spleen) Stage II = Involvement of two or more lymph node regions on the same side of the diaphragm Stage III = Involvement of lymph node regions or structures on both sides of the diaphragm

207

Stage IV = Involvement of extranodal site(s) beyond those designated as E, or > 1 extranodal deposit, or any involvement of liver or bone marrow _________________________________ A B No symptoms Unexplained weight loss > 10% during 6 months, unexplained fever > 38 C or recurrent night sweats during the previous month E Localized, solitary involvement of extralymphatic tissue, excluding liver and bone marrow _________________________________

Treatment and Prognosis

Therapeutic modalities and indications Indolent lymphomas may not be treated until definite clinical manifestations appear. Localized disease, involving only lymph nodes or an extra-nodal organ can be treated with RT. Advanced stages require chemotherapy with or without RT, depending upon the type of NHL, clinical manifestations, location, inter alia. Under investigation are new treatment regimens that include monoclonal antibodies against lymphoma cells, high dose chemotherapy and stem cell transplantation [2].
Follicular type Stage I or II is usually treated with RT. Alkylating agents, immunotherapy and radioimmunotherapy are used for Stage III or IV [1]. Highly responsive to chemotherapy and RT. Up to 25% of patients have spontaneous, though transient regression. With therapy 50 to 75% of patients achieve complete remission, although most will relapse. Over 20% will remain in remission for more than 10 years. Patients with follicular lymphoma often have histologic transformation to diffuse large B-cell lymphoma, heralded by rapidly developing lymphadenopathy. This is a poor prognostic scenario, but in some cases aggressive combined chemotherapy leads to complete remission [3]. Diffuse large B-cell Stage I or II Considered aggressive, and is treated with combined modalities: chemotherapy with doxorubicin and then RT. High-dose chemotherapy with stem cell rescue has been effective as salvage therapy for patients with diffuse large cell lymphomas [1]. A large number of patients with this type of lymphoma will be initially refractory to therapy or will relapse, and will be candidates for salvage therapy. Autologous bone marrow transplant seems to yield better results than does salvage chemotherapy [3]. Burkitts lymphoma Treated with intensive combination chemotherapy. Children and young adults have a 70% cure rate.

208

Small cell lymphocytic lymphoma Generally treated with chlorambucil or fludarabine. Gastric MALT lymphomas These lymphomas show a high response rate to H. Pylori eradication treatment, as mentioned earlier [1, 3]. Other primary extranodal lymphomas Additional therapy is needed for those at high risk for CNS failure, or involvement of contralateral paired organs [1]. Surgical resection of gastrointestinal lymphoma has been applied, with some controversy as to its role [18].

International prognostic index for NHL [3] 5 clinical risk factors: Age 60 or older Elevated serum LDH Low performance status (Karnofsky 70) > 1 site of extranodal involvement Ann Arbor stage III or IV. Patients with 0 or 1 factor: low risk, 5-year survival = 73% Patients with 2 factors: low-intermediate risk, 5-year survival = 51% Patients with 3 factors: high-intermediate risk, 5-year survival = 43% Patients with 4 or 5 factors: high risk, 5-year survival = 26%. __________________________________________________ Patients with a high prognostic index are more likely to need early therapy [3]. The prognosis varies very widely from a few weeks for the most aggressive tumors, to an indolent course without therapy for years or even for decades [2]. The overall 5-year survival rates in the early 1990s were about 50%, and among children nearly 80% [2].

12.2 Anatomical and Functional Imaging

12.2.1 FDG-PET and CT Functional imaging with FDG-PET scanning has been found to provide superior diagnostic accuracy compared to anatomical imaging using CT for staging and re-staging lymphoma. In their meta-analysis, Czernin and Phelps [19] showed a markedly improved specificity, which using CT was 39% (95% CI = 30 47) compared to 91% (95% CI = 86-96%)

209

for FDG-PET. Sensitivity of CT alone was 85% (95% CI = 78 91) and for FDG-PET 93% (95% CI = 89 98). Moreover, FDG-PET also reflected changes in tumor glucose metabolism associated with chemotherapy and this information was valuable for assessing prognosis. Assessment of Response to Therapy Weihrauch et al. [20] showed in their review that post-treatment evaluation of lymphomas with FDG-PET provided positive predictive values ranging from 57 to 100%, and negative predictive values ranging from 67 to 100%. A positive PET scan was more often indicative of progressive disease for NHL than for Hodgkins lymphoma. Spaepen et al. [21] report that among 70 patients with newly diagnosed aggressive NHL, a mid-treatment FDG-PET scan provided stronger prognostic information than did the International Prognostic Index. They suggest that an early restaging FDG-PET scan might be used to tailor induction chemotherapy in patients with aggressive NHL (p.
1356).

12.2.2 MRI MRI may be more sensitive than CT for evaluation of abdominal lymph nodes. MRI is also considered superior to CT for the investigation of bone marrow involvement by lymphoma [4, 22]. Pancreatic lymphomas are generally iso-intense to normal pancreas on both T1 and T2 weighted images [23]. Pancreatic adenocarcinomas are also generally difficult to visualize on MRI, making this important distinction difficult, unless vascular encasement of the latter can be delineated (see Chapter 15). 12.2.3 MRS assessment of lymph node involvement and response to therapy Proton MRS Schwarz et al. [24] examined 13 patients with extracranial lymphoma or germ cell tumors, including 5 patients with NHL using proton MRS3 immediately before chemotherapy. In order to enter the study, the tumor was required to be at least 3 x 3 cm if superficial, or 5 x 5 cm if deeper lying. The 3.2 ppm resonance was assigned to the trimethyl ammonium N-(CH3)3 moiety of total choline. Signals due to mobile
3 Schwarz, et al. [24], used 1.5T, single voxel, STEAM, TR=2000 ms, TE=20 ms. Follow-up scans attempt to be at the same site as the first. Used MRUI/VARPRO. Gaussian model functions fitted, metabolite signal values expressed as percentages of the water signal at the same TE.

210

lipids ([CH2])n at 1.3 ppm and CH3 at 0.9 ppm were seen in most spectra; however, many of these tumors were surrounded by fat. These authors assessed changes in the total choline to water ratio in the first post-treatment scan as a fraction of the pre-treatment ratio in 3 of the 5 patients with NHL shown in Table 12.1, as well as in 6 of the other patients (with Hodgkins lymphoma or teratomas). Overall, these authors [24] observed in the 9 patients with pre-and posttreatment spectral information, the changes fell into two clearly distinct patterns that correlated with subsequent clinical response (p. 962). In 7 patients the total choline to water ratio fell at the posttreatment scan (taken between 5 and 37 days after initiation of therapy). All of these patients were later judged to have a partial response to therapy at 54 to 93 days. In the other 2 patients, the total choline to water ratio remained approximately constant, and both of these patients were considered to have progressive disease. Importantly, the reduction in total choline to water ratio was observed earlier than volume changes assessed using MRI. The authors of Ref. [24] note that the proton MRS spectra appeared broadly similar across the different tumor types and histopathology {and that} any metabolite signals in the 2-2.5 ppm region of the spectrum are likely to be confounded by co-resonant lipid signals(p. 964). However, the resonances at 3.6 to 4.0 ppm were not assigned.

Table 12.1 Proton MRS Performed in Patients with NHL pre- & post-therapy
(From data of Schwartz et al. [24])
Post/Pre change-total choline: water ratio

Diagnosis Mantle cell

Site of Target Lesion Inguinal Node

Volume from MRI Post/Pre 1.02

Clinical Response Partial

0.31

Diffuse Large B-cell

Supraclavicular Fossa

Not available

Partial

Follicular

Mesentery

0.97

1.16

Progressive Disease

211

The authors [24] recognize that their patient group is small and heterogeneous, such that larger prospective series with specific pathologies need to be studied. They do note that their findings at 3.2 ppm could be from glycerophosphocholine, phosphocholine and choline, as well as possibly phosphoethanolamine, which has a resonance at about 3.2 ppm on the proton spectrum, as reported in studies using 31P MRS on lymphomas (see next subsection). Nevertheless, their general conclusion is that based on the differential changes in signal intensities in the {total choline} region of the spectrum following chemotherapy, we hypothesize that metabolic changes observable by 1H MRS may be predictive of subsequent clinical response. In this respect, changes are consistent with those observed in the phosphomonoester region of 31P spectra, which consists primarily of {phosphoethanolamine} PE and {phosphocholine} PC. In combination with other prognostic factors, such as the International Prognostic Index of lymphoma, biochemical information from MRS may find a role in individual patient management (p. 965). As discussed in Chapter 3, the advantages of proton MRS include being more easily available on standard MRI equipment, plus the better spatial resolution and sensitivity. In particular, MRS could be helpful for Phase I studies of new therapies against aggressive lymphomas, that target biological end points, but are not expected to dramatically alter tumor size. The observation of different trends in responders and non-responders suggests that further development of this method may provide a sensitive early indicator of metabolic response to treatment (p. 965) [24].
31

P MRS

Most studies of lymphomas have applied 31P MRS. An early investigation by Vogl et al. [25] of 15 patients with superficial masses (lymphomas as well as sarcomas, carcinomas, adenomas and tuberculosis) revealed that tumor growth was associated with increased PME concentrations. Subsequently, Negendank et al. [26] examined 21 patients with biopsy-proven NHL, 13 of whom were newly diagnosed and previously untreated, and 8 had recurrent disease after therapy. Seven of the patients had low grade, 11 had intermediate and 3 had high histopathologic grade. There were 2 patients with Stage I disease, 4 with Stage II, 6 with stage III and 9 with stage IV disease. Cell types included follicular and diffuse, predominantly; all but 2 of the patients had B-cell lymphomas. These authors applied 1H-decoupling and nuclear Overhauser enhancement to improve the resolution of 31P MRS localized to lymphoma-containing lymph nodes or masses within 10 cm of the body surface. All the MR spectra showed large PME signals (26% of total phosphorus) with a high phosphoethanolamine to

212

phosphocholine ratio. Glycerophosphoethanolamine and glycerophosphocholine were not detected, but there was a broad signal from membrane phospholipids in the phosphodiester region, comprising about 20% of the phosphorus. The nucleotide triphosphate (NTP) was prominent and inorganic phosphate was low, suggesting that the tissues were well perfused and that the cells were viable. These findings were similar in all grades and stages of NHL. The authors [26] interpreted these findings in light of in vitro studies, which differed from the present ones. They suggest: the pattern of phospholipid metabolites observed in NHL in vivo is partly a manifestation of sustained activation of phospholipase C or D (p. 3286). More recently, Griffiths et al. [27] in a multi-center study evaluated changes in PME/NTP ratios in relation to response to treatment in superficial lymph nodes of patients with NHL. There was a clear relation between change in this ratio and level of response: The 14 complete responders showed a highly significant fall in the PME/NTP ratio from 1.47 0.11 to 0.47 0.11, after therapy (p < 0.001), the 13 partial responders had a smaller, but still significant fall (p < 0.05) from 1.88 0.15 to 1.30 0.22, while the 16 patients who did not respond to therapy had a non-significant rise in PME/NTP ratio. Their results are graphically displayed in Figure 12.1.

Figure 12.1: Pre- and Post Treatment PME/NTP ratios (superficial lymph node) & Clinical response in 43 Patients with Non-Hodgkins Lymphoma (From data of Griffiths et al. [27])
3

2,5

2 PME/NTP

Pre-Treatment 1,5 Post-Treatment

0,5

0 Complete Responders Partial Responders Non-Responders

213

12.2.4

31

P-MRS assessment of hepatic lymphoma

Heindel et al. [28] compared the in vivo 31P MR spectra from liver of 13 patients with suspected hepatic involvement from Hodgkins lymphoma, with those from 22 healthy volunteers. Their findings were: Increased phosphomonoester to -NTP ratio in all the patients compared to the referents, Higher phosphomonoester to -NTP ratios and lower pH in patients with liver infiltration than those patients without hepatic involvement, Increased inorganic phosphate to -NTP ratios after therapy compared to prior to treatment with cytostatics, as examined in 3 patients. The authors [28] conclude that 31P-MRS can provide insights concerning hepatic involvement in patients with lymphoma, which are rarely gleaned from other imaging modalities. Dixon [29] reported that of 11 patients with Hodgkins and nonHodgkins lymphoma involving the liver, six showed elevated PME/Pi ratios on 31P MRS of the liver. In two of these patients this ratio dropped to within the normal range with clinical remission, whereas four patients in whom this ratio remained high after chemotherapy died of progressive disease. 12.2.5
31

P-MRS assessment of testicular lymphoma

Kiricuta et al. [30] presented a case report of a patient with testicular involvement from recurrent NHL, treated with mega voltage RT. Prior to therapy the 31P MR spectrum showed large phosphomonoester and phosphodiester peaks that overlapped the inorganic phosphate peak. After about half the RT administration the inorganic phosphate peak appeared and the pH lowered to 7.08 (close to normal in the testis of 7.02). After completion of RT the tumor disappeared and the patient went into complete remission. The authors suggest that inorganic phosphate and phosphomonoester might be used as markers of the response to RT.

References
[1] J. Coffey, D.C. Hodgson, M.K. Gospodarowicz, Therapy of non-Hodgkins lymphoma, Eur. J. Nucl. Med. Molecular Imaging 30 (Suppl. 1), S28-S36 (2003).

214

[2] M. Melbye, D. Trichopoulos, Non-Hodgkins lymphoma, in: H-O. Adami, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, Oxford, 2002, p. 535555. [3] J.O. Armitage, D.L. Longo, Malignancies of lymphoid cells, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 715-727. [4] J. O. Armitage, P.M. Mauch, N.L. Harris, P. Bierman, Non-Hodgkins lymphomas, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 2256-2316. [5] B. Alpen, J. Robbecke, T. Wundisch, M. Stolte, A. Neubauer, Helicobacter pylori eradication therapy in gastric high-grade non-Hodgkins lymphoma, Ann. Hematol. 80 (Suppl. 3), B106-B107 (2001). [6] J.A. Bukowski, W.W. Huebner, A.R. Schnatter, N.C. Wojcik, An analysis of the risk of Blymphocyte malignancies in industrial cohorts, J. Toxicol. Environ. Health A 66, 581-597 (2003). [7] E. Lee, C.A. Burnett, N. Lalich, L.L. Cameron, J.P. Sestito, Proportionate mortality of crop and livestock farmers in the United States, 1984-1993, Am. J. Indust. Med. 42, 410-420 (2002). [8] J. Simpson, E. Roman, G. Law, B. Pannett, Womens occupations and cancer: preliminary analysis of cancer registrations in England and Wales, 1971-1990, Am. J. Indust. Med. 36, 172-185 (1999). [9] T. Zheng, A. Blair, Y. Zhang, D.D. Weisenburger, S.H. Zahm, Occupation and risk of nonHodgkins lymphoma and chronic lymphocytic leukemia, J. Occup. Environ. Med. 44, 469-474 (2002). [10] M.I. Cano, M. Pollan, Non-Hodgkins lymphomas and occupation in Sweden, Int. Arch. Occup. Environ. Health 74, 443-449 (2001). [11] J.C. Schroeder, A.F. Olshan, R. Baric, et al., Agricultural risk factors for t (14; 18) subtypes of non-Hodgkins lymphoma, Epidemiology. 12, 701-709 (2001). [12] S. Roulland, P. Lebailly, Y. Lecluse, M. Briand, D. Pottier, P. Gauduchon, Characterization of the t (14; 18) BCL2-IGH translocation in farmers occupationally exposed to pesticides, Cancer Res. 64, 2264-2269 (2004). [13] A.S. Costantini, L. Miligi, D. Kriebel, et al., A multi-center case-control study in Italy on hematolymphopoietic neoplasms and occupation, Epidemiology 12, 78-87 (2001). [14] Y. Mao, J. Hu, A.M. Ugnat, K. White, Non-Hodgkins lymphoma and occupational exposure to chemicals in Canada, Ann. Oncol. 11 (Suppl. 1), 69-73 (2000). [15] L. Varoczy, L. Gergely, M. Zeher, G. Szegedi, A. Illes, Malignant lymphoma-associated autoimmune diseasesa descriptive epidemiological study, Rhematol. Int. 22, 233-237 (2002). [16] D.J. van Spronsen, M.L. Janssen-Heijnen, W.P. Breed, J.W. Coebergh, Prevalence of comorbidity and its relationship to treatment among unselected patients with Hodgkins disease and non-Hodgkins lymphoma, Ann. Hematol. 78, 315-319 (1999). [17] K. Belkic, P.L. Schnall, C. Savic, P.A. Landsbergis, Multiple exposures: Towards a model of total occupational burden, in: P.L. Schnall, K. Belkic, P.A. Landsbergis, Baker D (eds.) Occupational Medicine: State of the Art Review. The Workplace and Cardiovascular Disease. 15, 94-105 (2000). [18] M.L.V. Fasan, E. Morandi, P. Fociani, et al., AIDS-associated gastrointestinal lymphoma: Is there a role for surgery in the standard of care? J. Acquir. Immune Defic. Syndr. 34, 345-347 (2003).

215

[19] J. Czernin, M.E. Phelps, Positron emission tomography scanning: Current and future applications, Annu. Rev. Med. 53, 89-112 (2002). [20] M.R. Weihrauch, M. Dietlein, H. Schicha, V. Diehl, H. Tesch, Prognostic significance of 18F-flurodeoxyglucose positron emission tomography in lymphoma, Leukemia Lymphoma, 44, 15-22 (2003). [21] K. Spaepen, S. Stroobants, P. Dupont, et al., Early restaging positron emission tomography with (18) F-fluorodeoxyglucose predicts outcome in patients with aggressive non-Hodgkins lymphoma, Ann. Oncol. 13, 1356-1363 (2002). [22] V. Diehl, P.M. Mauch, N.L. Harris, Hodgkins disease, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p.2339-2387. [23] M.K. Kalra, M.M. Maher, P.R. Mueller, S. Saini, State-of-the-art imaging of pancreatic neoplasms, Br. J. Radiol. 76, 857-865 (2003). [24] A.J. Schwarz, N.R. Maisey, D.J. Collins, D. Cunningham, R. Huddart, M.O. Leach, Early in vivo detection of metabolic response: a pilots study of 1H MR spectroscopy in extracranial lymphoma and germ cell tumors, Br. J. Radiol. 75, 959-966 (2002). [25] T. Vogl, F. Peer, H. Schedel, et al., 31P-spectroscopy of head and neck tumorssurface coil technique, Magn. Reson. Imaging 7, 425-435 (1989). [26] W. G. Negendank, K.A. Padavic-Shaller, C-W. Li, et al., Metabolic characterization of nonHodgkins lymphomas in vivo with proton-decoupled phosphorus MRS, Cancer Res. 55, 32863294 (1995). [27] J.R. Griffiths, A.R. Tate, F.A. Howe, M. Stubbs, as part of the Multi-Institutional Group on MRS Application to Cancer, Magnetic resonance spectroscopy of cancer - practicalities of multicentre trials and early results in non-Hodgkins lymphoma, Eur. J. Cancer 38, 2085-2093 (2002). [28] W. Heindel, R. du Mesnil de Rochemont, H. Kugel, et al. 31P-MR spectroscopy of the human liverthe spectral indications of lymphoma infiltration, ROFO-Frotschritte 167, 62-70 (1997). [29] R.M. Dixon, NMR studies of phospholipid metabolism in hepatic lymphoma, NMR Biomed. 11, 370-379 (1998). [30] I. C. Kiricuta, R.G. Bluemm, J. Ruhl, H.K. Bever, 31P-MR spectroscopy and MRI of a testicular non-Hodgkin lymphoma recurrence to monitor response to irradiation. A case report, Strahlenth. Onkol. 170, 359-364 (1994).

217

Chapter 13

SarcomasMusculoskeletal Tumors
_______________________________________________________________________________

Sarcomas are primary tumors of bone and soft tissue (muscles, tendons, fat, fibrous tissue, synovial tissue, vessels and nerves); these are mainly of mesodermal origin, but occasionally arise from neurectoderm [1].

13.1 Overview of epidemiological & clinical aspects

Incidence and prevalence/morbidity and mortality Sarcomas comprise less than 1% of all cancers and occur in all age groups. They are, however, among the most frequently occurring solid tumors in children and represent the fifth most common cause of cancer-related deaths among children. Soft tissue sarcomas are more common than those affecting bone [1]. Etiology/risk factors Etiologic research is considered difficult because of the rarity of these tumors. Inherited susceptibility Genetic predisposition (autosomal dominant) is seen in about 8-9% of children with soft tissue sarcomas [2]. Li et al. [3] assessed parental cancer as a risk factor for bone malignancies, using the Swedish Family-Cancer Database. They also investigated the risk of second malignancy after childhood bone cancer. Most bone cancer cases were found to be sporadic. However, there was an increase in osteosarcoma in the offspring (up to age 25) of those with breast and prostate cancers

218

SIR = 1.7). This was attributed to Li-Fraumeni syndrome. Giant cell sarcoma showed an SIR = 2.9 with parental breast cancer. For earlyonset chondrosarcoma, an SIR = 6.8 was reported in association with parental renal cancers. There was an increased risk of second bone and connective tissue cancers after childhood bone cancer. Congenital immunodeficiency, as well as the following genetic disorders, is associated with an increased risk of sarcoma [1 2]:
Li-Fraumeni syndrome Familial cancer syndrome with germ-line mutations in p53, associated with soft tissue and osteogenic sarcoma. Neurofibromatosis I May occasionally undergo malignant transformation. Inherited retinoblastoma Associated with soft tissue and osteogenic sarcoma. Breaks in chromosome 18 and on X Associated with synovial sarcomas in 90% of cases. Gardners syndrome (familial adenomatous polyps) Associated with fibrosarcoma, desmoid tumors. Werners syndrome Associated with soft tissue sarcomas. Gorlins syndrome (nevoid basal cell carcinoma syndrome) Associated with fibrosarcoma, rhabdomyosarcoma. Tuberous sclerosis Associated with rhabdomyosarcoma.

Environmental/Occupational Factors Exposure to ionizing radiation is well recognized as a risk factor for soft tissue and bone sarcomas. There is some suggestion of a link between phenoxy herbicide exposure in forestry, farming and railroad workers, but there are also null findings. Cohort data have linked 2,3,7,8-tetrachrorodibenzo-p-dioxin exposure among industrial workers, and risk of sarcoma, but null findings are also reported [2]. Pukkala et al. [4] examined cancer incidence among 2 269 male world-class athletes from 1967 1995. Mainly the risk of cancer was low. However, among hurdlers there was a significantly increased

219

risk of bone and soft tissue sarcomas. The authors suggest that this might be related to injuries during the active sport period (p. 216).

Medical Conditions and Related Exposures

Acquired immunodeficiency Increases the risk of sarcoma. Kaposis sarcoma has a viral etiology (HIV type I, human herpes virus), but this is not the case for other soft tissue sarcomas. RT and chemotherapy (alkylating agents) Radiation therapy is a risk factor for soft tissue and bone sarcomas. These usually occur within the irradiated field, and the risk increases with time. The ensuing sarcomas are usually high-grade and have a poor prognosis [5]. Patients with retinoblastoma are found to be at a dose-dependent increased risk for a second primary tumor in relation to RT [1, 2]. Hawkins et al. [6] used the National Childhood Tumor Registry of the U.K. in a nested case-control study of 59 patients with a second primary bone cancer and 220 referents, all with a primary childhood cancer. The risk of bone cancer increased significantly (p < 0.001) with increased cumulative dose of radiation to the bone. With < 10 Gy, the risk was non-significant. In that study [6] there was also an increase in bone cancer (p = 0.04, one-tailed) for increased cumulative dose of alkylating agents. The authors conclude: this populationbased study provides grounds for reassurance of the majority of survivors in that their risk of developing bone cancer within 20 years of 3-year survival did not exceed 0.9%. The higher risks found for bone cancer following the other specific rare types of childhood cancer provide a rational basis of surveillance
(p. 270).

This excess incidence of sarcomas is mainly seen in heavily irradiated tissues [7, 8]. Hall and Wuu [8] note that with a transition from conventional RT to 3D-conformal RT there is a reduction in the volume of normal tissues which received high dose of radiation, and suggest that this should decrease the number of induced sarcomas. In contrast, however, by involving more fields, IMRT exposes a larger volume of normal tissue to low dose radiation. There is also an increased total body exposure due to leakage radiation. These two factors will tend to increase the risk of second cancers compared to conventional RT [8]. Hall [7] points out the importance of these findings for young patients, and, in particular, underscores the need to identify by predictive assays radiosensitive patients who would be at risk for late sequelae. Unfortunately, however, these assays have not been reliable predictors, and the author underscores the need for identifying gene mutations that increase radiosensitivity. One example is the rare ataxia-telangiectasia, which, as noted, confers very high radiosensitivity. Another mutation is of the BRCA 1 and BRCA 2 [7]. Benign bone tumors These can undergo malignant transformation. Sarcomas can also arise from soft tissues (e.g. extraskeletal osteosarcoma).

220

Chronic lymphedema post-radical mastectomy (Stewart-Treves syndrome) Well recognized as a risk factorpost-mastectomy post-irradiated lymphedematous arm [9]. This is not considered to be radiation-induced because the sarcomas develop outside the irradiated field [2]. Childhood cancer There is an increased risk of second bone and connective tissue cancers after childhood bone cancer, independent of radiation therapy. Neglia et al. [10] analyzed the U.S. and Canadian Childhood Cancer Survivor Study cohort of 13 581 children diagnosed with cancer before age 21 and surviving at least 5 years. There were 314 subsequent malignancies among 298 patients. The largest excess was for bone cancers (SIR = 19.14). Overall, there were 1.88 excess cancers per 1000 years of patient follow-up. The authors emphasize the importance of surveillance and early detection. Inguinal hernias Valery et al. [11] performed a case-control study among the non-farming population of Australia; included were 106 patients with Ewings sarcoma and 344 population-based referents. A significantly increased risk was found for having had an inguinal hernia repair (OR = 5.6 (95% CI = 1.3 - 6.4)).

Clinical Presentation Soft tissue sarcomas Most frequently present as an asymptomatic mass. They can also cause mechanical symptoms due to nerve entrapment, pressure, etc. 50% of soft tissue sarcomas occur in the extremities, two-thirds of which are in the lower extremities.

Osteosarcomas Sixty percent occur in children and adolescents. These have a predilection for long bone metaphyses; in adults these tumors are often secondary to RT. Osteosarcomas typically present with pain and swelling in the affected area. Chondrosarcoma These comprise 20-25% of all bone sarcomas and are mainly seen in adults. Chondrosarcomas have a predilection for flat bones, particularly pelvic and shoulder girdles, but also occur in the diaphyseal portion of long bones. They tend to be indolent, can arise from enchondromas or osteochondromas, but also occur spontaneously. New-onset of pain, inflammation and increased size are suggestive manifestations.

221

Ewings sarcomas These account for 10-15% of all bone sarcomas, and are most common among adolescents, affecting most often the diaphyseal region of long bones as well as flat bones.
Differential Diagnosis For Sarcomas of bone
Bony metastases Especially from prostate, breast, lung, kidney, bladder, thyroid, lymphomas and multiple myeloma. These can produce osteoblastic (detected with radionuclide scans, and showing a high alkaline phosphatase, and possibly hypocalcemia) or osteolytic lesions. Osteoporosis In women, a clue is the preservation of cortical bone [1].

Soft tissue sarcomas

Benign soft-tissue lesions, Metastatic carcinoma, melanoma, lymphoma.

Classification and Grading There are many different types of soft tissue sarcomas, including rhabdomyosarcoma (aggressive), leiomyosarcoma, and liposarcoma. These typically metastasize via the blood, with the lung being the most common site. Staging is done by size, grade, node status and metastases [1], as follows:

Stage I = Well to moderately differentiated (G1-2), T1 ( 5 cm), without nodal involvement (N0) or metastases (M0). Stage II = Tumor > 5cm and G1-2 or Tumor 5 cm, but poorly differentiated, without nodal involvement (N0) or metastases (M0). Stage III = Tumor > 5cm and poorly differentiated, without nodal involvement (N0) or metastases (M0). Stage IV = Nodal involvement (N1) or metastases (M1).

222

Treatment and Prognosis

Surgery Defined as successful if a wide excision is made, with a negative margin. Stage I disease is usually treated with surgery only; 5-year survival is 98.8 %. Surgery is the mainstay of therapy for chondrosarcomas; these are usually resistant to chemotherapy (with the exception of dedifferentiated chondrosarcoma and mesenchymal chondrosarcoma, which are treated with chemotherapy) [1].

Radiation Therapy Used as adjuvant to limb-sparing surgery. Chemotherapy This is the mainstay for Ewings sarcoma, rhabdomyosarcomas, and peripheral neuroepithelial tumors. Doxorubicin has shown significant improvement in local control and disease-free survival, but overall survival is only improved for sarcomas of the extremity. Applying neoadjuvant chemotherapy, and continuing post-operatively only if the patient responds pre-operatively has been a strategy used to help spare non-responsive patients from the toxicity of chemotherapy. Osteosarcomas are usually high-grade; response to chemotherapy is the best prognostic indicator. The standard approach is neoadjuvant therapy, followed by surgery (limb-sparing in 80% of patients), and then post-operative chemotherapy. Long-term survival is 60-80% in osteosarcomas of the extremity. These are radio-resistant tumors; RT is not routinely applied [1]. Stage IV disease is usually incurable, but up to 20% of patients who respond to therapy (surgical resection of metastases plus chemotherapy) are long-term survivors [1]. In patients with unresectable soft tissue sarcoma of the extremities, isolated limb perfusion has been found to produce substantial tumor regression, often enabling limb-sparing resection [12].
Summarizing the current situation with respect to therapy and outcomes in soft-tissue sarcomas, Cormier and Pollock [13] state: despite improvements in local control rates with wide local resections and radiation therapy, metastasis and death remain a significant problem in 50% of patients who present with high risk soft tissue sarcomas ... Progress in the molecular characteristics of these tumors should in the near future translate into molecularly based therapies that can be incorporated into standard treatment strategies (p. 94).

223

13.2 Current approach to 1 diagnosis & staging

Plain-film x-ray, MRI and nuclear medicine procedures are complementary modalities in the diagnosis and staging of musculoskeletal tumors. Plain films generally show the lesion and may provide some indication of its aggressiveness [14]. Pre-operative evaluation
Osteosarcomas

Appear typically on plain-film x-ray as a destructive lesion with a moth-eaten appearing, spiculated periosteal reaction and cuff of new bone (Codmans triangle). CT scan defines bony destruction and calcification pattern. MRI is superior for defining the intramedullary and soft tissue extension. T1 weighted images help differentiate these lesions from osteoporosis [14]. Bone scans are used to detect bony metastases, with CT and chest x-ray for lung metastases. Angiography has been used for assessing pre-operative response to chemotherapy [1] and demonstrates vascular patterns for limb sparing procedures [15].
Chondrosarcomas

Difficult to distinguish from benign lesions. Their typical radiographic pattern is lobular with mottled, punctate or annular calcification of the cartilage [1].
Ewings sarcomas

Often show an onion peel periosteal reaction on plain films with a large soft tissue mass best seen with CT or MRI. The presence of p30/p32 from the mic-2-gene is a cell-surface marker [1].
Soft tissue tumors

Best evaluated on MRI, which provides excellent anatomic detail and can often distinguish benign from malignant lesions. A thin-rimmed homogeneous, high intensity area on T2 weighted images is considered suggestive of a simple cyst, whereas a lesion of heterogeneous high intensity is more indicative of a solid malignancy. However, these findings are frequently inconclusive, in which case MRI is used to indicate where a biopsy should be taken. On the other hand, MRI can yield a definitive diagnosis for lipomas, which appear as homogeneous, high intensity lesions on T1 weighted imagesthis is chemically specific for lipomas due to the presence of macroscopic fat, such that a biopsy is not needed when the MRI findings are unequivocal [14].

224

FDG-PET has also been used to identify metastases. Biopsy for definitive diagnosis Bone biopsy must be properly executed and this requires high skill and care; most frequently the placement is improper. Precautions are needed to avoid contamination when performing an open biopsy, but tumor cells contaminate the transversed tissue planes and compartments. Therefore, the biopsy sites must be removed en bloc [15]. Post-operative evaluation
MRI is especially used for monitoring neoadjuvant chemotherapy, but

cannot distinguish tumor, hemorrhage, necrosis and edema on T2 weighted images, nor tumor from inflammation on T1 weighted images.
Thallium scintigraphy has also been used for following response

to therapy.
FDG-PET

This is under investigation for assessing response to therapy [15].

13.3
1

MRS in the assessment of sarcomas

13.3.1 Primary detection H MRS

Studies using in vivo proton MRS to assess sarcomas are sparse, and fairly inconclusive. Schick et al. [16] performed proton MRS and MRSI1 in 37 examinations of 27 patients with bone tumors (various sarcomas plus 14 benign tumors). Resonances from fat were found in osteochondromas but not in the malignant tumors. Resonances with J coupling were found in 7 of the 13 malignant tumors. Oya [17] et al. performed 49 proton MRS studies2 of bone and soft tissue tumors of various types. These authors report an unassigned resonance at 2.0-2.1 ppm in 6 of the 47 lesions (clear cell sarcoma (2/2), Ewing sarcoma (1/1), malignant fibrous histiocytoma (1/3), malignant Schwannoma (1/1) and mucoepidermoid carcinoma (1/1)). This is interpreted to be suggestive of an untreated cancer of neuroectodermal origin.

1 2

Schick et al. [16] used 1.5T, single voxel and MRSI, PRESS, TE 50ms. Oya et al [17] used 1.5T, PRESS, TR=1500 or 2000 ms, TE=30, 60, 136 and 272 ms.

225

31

P MRS

There are more clinical investigations using in vivo 31P MRS. Some of these are helpful for primary diagnosis of sarcomas. In an early study Lenkinski et al. [18] examined 7 patients with bone tumors using MRI and 31P MRS. They observed that the bone tumors were surrounded by muscle tissue that contained high concentrations of phosphocreatine (PCr), and therefore recommend assessing ratios with -NTP rather than PCr. The PME/-NTP ratio was increased compared to normal muscle in all of the examined bone tumors. Subsequently, Li et al. [19] applied proton decoupling and nuclear Overhauser-enhancement to obtain well resolved 31P MRSI of 20 soft tissue sarcomas. Fifteen of these 20 examinations showed high phosphoethanolamine but not glycerophosphoethanolamine. Glycerophosphocholine was observed in only 4 cases. High NTP and low Pi suggested a large percentage of viable cells, and this was confirmed histologically in 13 of the sarcomas. The same year, a paper was published by Negendank et al. [20] using 31P MRS with proton-decoupling and nuclear Overhauserenhancement in 53 patients with cancers of various types (including 17 sarcomas and 2 metastatic adenocarcinomas to bone). These cancers all shared the following characteristics: Phosphoethanolamine was the predominant phosphomonoester, Glycerophosphoethanolamine and glycerophosphocholine were rarely detected, A broad phosphodiester resonance was found, most likely from membrane phospholipid, Prominent NTP. Kettelhack et al. [12] note that several groups have shown tumors to be clearly distinguished from normal muscle and soft tissue by PME, PDE and Pi.

13.3.2

31

P MRS to assess response to therapy

A 1995 paper by Negendank [21] provides a review of the published investigations of sarcomas using 31P MRS, stating that despite technical limitations (poor localization to regions to interest leading to severe contamination from muscle and poor resolution) that many of the studies provide clinically important insights. In particular, Negendank noted that this methodology could be helpful for predicting response to therapy. Application of proton decoupling and nuclear Overhauserenhancement together with dual-tuned surface coils and accurate

226

localization to regions of interest using 3D MRSI have helped overcome the above-cited difficulties. Just after Negendanks review, a paper was published by Moller et al. [22] in which 28 patients with musculoskeletal cancers were examined using 31P MRS. Prior to treatment, PME and PDE were high, Pi moderately increased and PCr was low. The intracellular pH was slightly alkaline. At follow-up, energy-rich phosphates decreased in the non-responders, whereas PME, Pi and often PDE increased. In 5 of the patients who responded to chemotherapy, initially there was a pattern of ischemic cell injury followed by a phase suggesting tumor activation. Suggestive of tumor response were: (PCr + Pi) / total phosphate 0.35 Pre-treatment ratio of (PCr + Pi) / -NTP 1.5 Accelerated increase in total low-E phosphates / total high-E phosphates after initiation of chemotherapy Long term decrease in total low-E phosphates/ total high-E phosphates. The authors [22] conclude: such spectroscopic predictors for treatment response proved to be superior to currently used indices such as tumor size (p. 347). In a 1998 paper Sijens [23] reviewed the application of 31P MRS for cancers of the human extremities, and evaluated these in relation to data on murine sarcomas. The author notes that there is low spatial resolution with 1.5 or 2T, and that this is a major limitation, stating: There are early spectral changes in human extremity sarcomas monitoring after therapy 31P MR spectra measured before treatment and the changes in phosphate metabolites measured shortly thereafter, correlate with the clinical response after 2 or 3 months. Larger clinical studies are needed to confirm whether correlations of, for instance, pretreatment tumor pH with necrosis at resection and Pi decrease with tumor regression, can be used as a predictive test for clinical response
(p. 341).

More recently, Kettelhack et al. [12] performed 31P MRS3 among 32 patients with locally advanced, unresectable soft tissue tumors (28 sarcomas, 4 melanomas), prior to and after isolated limb perfusion (ILP) with tumor necrosis factor- and melphalan or cytostatics. According to clinical criteria, 15 of the 32 patients showed a partial response, another 15 showed no change and 2 had progressive disease. As shown in Figure 13.1, the mean PME to -ATP ratios
3

Kettelhack et al. [12] used 1.5T, 31P MRS, TR=2000ms, linear phase correction was applied manually, least squared fit assuming Lorentzian line shape, peak ratios were calculated, pH was calculated based on the difference between fitted line positions of PCR and Pi.

227

decreased after ILP compared to before ILP in all groups, but this decrease was significantly greater among the partial responders compared to the non-responders (p = 0.02). Figure 13.1: PME /-ATP ratios after vs prior to isolated limb perfusion for soft tissues, according to clinical response (From data of Kettelhack et al. [12])
80% 70% 60% 50% 40% 30% 20% 10% 0% Partial Response No Change Disease Progression

The authors reported a substantial difference between the clinical response parameters (WHO criteria) versus histological response. A total of 17 patients showed a histological response (necrosis 90%), 7 of these patients showed no clinical response. Combining histological and clinical response, 31P MRS had 94% specificity and 68% sensitivity for predicting response. The authors conclude: utilizing changes in PME/-ATP ratios, 31P MRS is a highly specific tool with which to predict histological response in this setting. This finding may be of major value in those patients in whom the decision to perform a major resection or amputation must be made for local tumor control (p. 1557).

13.3.3 In vitro MRS studies of sarcomas

31

P MRS

In the previously cited study, Li et al. [19] also performed highresolution in vitro 31P MRS on 4 of the sarcomas that they had studied in vivo. In addition, 6 other sarcomas were analysed to confirm the in vivo assignments. Seven high-grade sarcomas with pleomorphic or round cells rather than spindle cells, contained an unidentified phosphodiester resonance in vivo, but this was absent in the in vitro

228

analyses. This was interpreted to possibly be an abnormally mobile membrane component.

1

H MRS

Liposarcoma Millis et al. [24] note that there are major controversies surrounding the prognosis of patients with liposarcoma. Currently this is based on the presence of lipoblasts, atypical adipocyte nuclei, and qualitative assessment of cellularity and cell size. In the present study, the authors performed HRMAS proton MRS on specimens from 30 patients with liposarcoma and 5 patients with lipoma, and found the following: MRS-visible triglycerides correlated with degree of liposarcoma differentiation, being 33-fold greater in well differentiated compared to myxoid/round cell liposarcoma. Moreover, the myxoid/round cell liposarcoma triglyceride level was 6-fold greater than dedifferentiated pleomorphic liposarcomas. MRS-visible phosphatidylcholine was used to estimate total tissue cell membrane phospholipid, and this was correlated with liposarcoma differentiation. Pleomorphic liposarcoma, which has the highest propensity to metastasis, had 3-fold greater MRS-visible phosphatidylcholine compared to the somewhat less aggressive dedifferentiated liposarcoma. MRS-visible phosphatidylcholine was two times higher in welldifferentiated liposarcoma compared to lipomas. MRS-visible phosphatidylcholine was three times higher in hypercellular myxoid/round cell liposarcomas compared to pure myxoid liposarcomas.

The authors [24] conclude: NMR-derived parameters of tissue lipid may be used for objective distinction of liposarcoma histologic subtype/grade and lipoma from liposarcoma. These biochemical parameters may ultimately improve prognostication in patients with liposarcoma (p. 257).
Using 1D and 2D HRMAS, Singer [25] states: the proposed relationship between tissue phosphatidylcholine content with tissue cellularity and the percentage of cells in S-phase is supported by the NMR analysis of phosphatidylcholine content in the various histological subtypes of liposarcoma. There was a 1.6-fold increase in phosphatidylcholine in the more cellular and mitotically active dedifferentiated liposarcoma as compared to the well-differentiated liposarcoma (p = 0.05). The pleomorphic liposarcomas, which on average had the highest mean mitotic activity and the highest rate of distant metastasis of all the histological types of liposarcoma, were found to have the highest NMR-visible tissue phosphatidylcholine levels. The average phosphatidylcholine level found in the pleomorphic liposarcoma was over

229

three times larger than the average phosphatidylcholine level measured in the dedifferentiated liposarcoma group (p = 0.01). Even within a given histologic cell type, there appeared to be a direct correlation between phosphatidylcholine content and cellularity (p. 19). In that study, Singer [25] examined the N-methyl region of phosphatidylcholine (3.19 to 3.34 ppm) as a ratio to triglycerides plus phospholipid and other metabolites (0.5 to 4.5 ppm). The myxoid/round cell liposarcomas had a fractional N-methylcholine ratio 37 times greater than the well-differentiated liposarcoma (p = 0.0001).

Singer [25] concludes: improvements in NMR technology have now enabled quantitative detection of lipid and phospholipid sarcoma cellularity, proliferation, and differentiation state. Future work needs to address whether this biochemical lipid analysis could serve as objective prognostic determinants for grading soft tissue sarcoma (p. 20).
D/C ratio as an index of UnsaturationCorrelates with Leiomyosarcoma Mitotic Activity In vitro NMR with quantification using 2D total correlation spectroscopy (TOCSY) was also performed in the laboratory of Singer [25], describing cross-peak C from methylene protons adjacent to only one unsaturated site proton, whereas cross-peak D is from the correlation between olefinic protons and adjacent diallylic methylene protons of the fatty acyl chain. The ratio of cross-peaks D/C is used as an NMR-visible measure of the degree of unsaturation. The author reports that the D/C ratio for high-grade leiomyosarcomas was significantly higher compared to low-grade leiomyosarcomas, and the latter is significantly higher than normal smooth muscle. The D/C cross-peak ratio is found to be proportional to mean mitotic activity.

Singer [25] has reviewed new diagnostic modalities in soft tissue sarcoma. He notes that the grade and size of the sarcoma are the most important factors for predicting the risk of relapse and overall survival, and points out that there is, however, substantial discordance, even among pathologists who are experts in this field. Evidence is increasing that the composition of membrane phospholipid in tumor tissue indicates cellularity, proliferative capacity and differentiation, with the most abundant phospholipid in human cell membranes being phosphatidylcholine. Singer [25] states: the level of NMR-visible phosphatidylcholine in a tissue may serve as an estimate of bulk membrane phospholipid content for cells that are in various phases of the cell cycle in that tissue with the majority of membrane phospholipid accumulating in the fraction of cells in S-phase. NMR estimates of total tissue phosphatidylcholine most likely are determined by the tissue cellularity (number of cells per unit volume) as well as the

230

number of cells per unit volume that are progressing through the synthesis phase of the cell cycle prior to cell division (p. 17).

References
[1] S. R. Patel, R.S. Benjamin, Soft tissue and bone sarcomas and bone metastases, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 625-628. [2] M.F. Brennan, K.M. Alektiar, R.G. Maki, Soft tissue sarcoma, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p.1841-1890. [3] X. Li, K. Hemminki, Parental cancer as a risk factor for bone cancer: a nation-wide study from Sweden, J. Clin. Epidemiol. 55, 111-114 (2002). [4] E. Pukkala, J. Kaprio, M. Koskenvuo, U. Kujala, S. Sarna, Cancer incidence among Finnish world class male athletes, Int. J. Sports Med. 21, 216-220 (2000). [5] S.R. Patel, Radiation-induced sarcoma, Curr. Treat.Opt. Oncology 1, 258-261 (2000). [6] M.M. Hawkins, L.M. Wilson, H.S. Burton, et al., Radiotherapy, alkylating agents and risk of bone cancer after childhood cancer, J. Natl. Cancer Inst. 88, 270-280 (1996). [7]. E.J. Hall, Do no harm. Normal tissue effects, Acta Oncologica 40, 913-916 (2001). [8] E.J. Hall, C.S. Wuu, Radiation-induced second cancers: the impact of 3D-CRT and IMRT, Int. J. Radiat. Oncol. Biol. Physics. 56, 83-88 (2003). [9] N. Penel, C. Nisse, S. Feddal, E. Lartigau, Epidemiology of soft tissue sarcomas in adults, Presse Med. 30, 1405-1413 (2001). [10] J.P. Neglia, D.L. Friedman, Y. Yasui, et al., Second malignant neoplasms in five-year survivors of childhood cancer: childhood cancer survivor study, J. Natl. Cancer Inst. 93, 618-629 (2001). [11] P.C. Valery, W. McWhirter, A. Sleigh, G. Williams, C. Bain, A national case-control study of Ewings sarcoma family of tumors in Australia, Int. J. Cancer 105, 825-830 (2003). [12] C. Kettelhack, M.V. Wickede, T. Vogl, U. Schneider, P. Hohenberger, 31Phosphorusmagnetic resonance spectroscopy to assess histologic tumor response noninvasively after isolated limb perfusion for soft tissue tumors, Cancer 94, 1557-1564 (2002). [13] J.N. Cormier, R.E. Pollock, Soft tissue sarcomas, Ca: Cancer J. for Clinicians 52, 94-109 (2004). [14] A.E. Li, D.A. Bluemke, Magnetic resonance imaging in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 669-679. [15] M.M. Malawer, M.P. Link, S.H. Donaldson, Sarcomas of bone, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 1891-1935. [16] F. Schick, S. H. Duda, O. Lutz, C.D. Claussen, Lipids in bone tumors assessed by magnetic resonance: chemical shift imaging and proton spectroscopy in vivo, Anticancer Res. 16, 1569-1574 (1996).

231

[17] N. Oya, J. Aoki, T. Shinozaki, H. Watanabe, K. Takagishi, K. Endo, Preliminary study of proton magnetic resonance spectroscopy in bone and soft tissue tumors: an unassigned signal at 2.0-2.1 ppm may be a possible indicator of malignant neuroectodermal tumor, Radiat. Med. 18, 193-198 (2000). [18] R.E. Lenkinski, J. Listerud, M.A. Shinkwin, et al., Magnetic resonance imaging and spectroscopy of bone tumors and bone marrow, Investig. Radiol. 24, 1006-1010 (1989). [19] C.W. Li, A.C. Kuesel, K.A. Padavic-Shaller, et al., Metabolic characterization of human soft tissue sarcomas in vivo and in vitro using proton-decoupled phosphorus magnetic resonance spectroscopy, Cancer Res. 56, 2964-2972 (1996). [20] W.Negendank, C-W. Li, K. Padavic-Shaller, J. Murphy-Boesch, T.R. Brown, Phospholipid metabolites in 1H-decoupled 31P MRS in vivo in human cancer: Implications for experimental models and clinical studies, Anticancer Res. 16, 1539-1544 (1996). [21] W.G. Negendank, MR spectroscopy of musculoskeletal soft-tissue tumors, Magn. Reson. Imaging Clin.N.Am. 3, 713-725 (1995). [22] H.E. Moller, P. Vermathen, E. Rummeny, et al., In vivo 31P spectroscopy of human musculoskeletal tumors as a measure of response to chemotherapy, NMR Biomed. 9, 347-358 (1996). [23] P.E. Sijens, Phosphorus MR spectroscopy in the treatment of human extremity sarcomas, NMR Biomed. 11, 341-353 (1998). [24] K. Millis, P. Weybright, N. Campbell, et al., Classification of human liposarcoma and lipoma using ex vivo proton NMR spectroscopy, Magn. Reson. Med. 41, 257-267 (1999). [25] S. Singer, New diagnostic modalities in soft tissue sarcoma, Semin. Surg. Oncol. 17, 11-22 (1999).

233

Chapter 14

Renal Cell Carcinomas

_______________________________________________________________________________

In adults approximately 80-90% of the malignant tumors of the kidney are renal cell carcinoma (RCC). Cancers of the renal pelvis are etiologically more similar to bladder cancer [1].

14.1 Overview of epidemiological & clinical aspects

Incidence and prevalence/morbidity and mortality The highest rates of RCC are in Europe, Australia, North America, intermediate in southern Europe and Japan, and low elsewhere in Asia, in Africa and the Pacific. These differences may be related, at least in part, to diagnostic intensity, as may the rising trends. The incidence in the U.S. is higher among African-Americans than the white population. In 2000, there were an estimated 35 000 new cases in North America and 45 000 in the E.U. [1]. Peak incidence is seen between the ages of 50 and 70, and the male: female ratio is about 2:1. Renal cell carcinoma accounted in the year 2000 for 1.5% of cancer deaths. Mortality rates parallel incidence rates [1]. Etiology/risk factors Inherited susceptibility: Although most cases of renal cell carcinoma are sporadic [2], there are a number of inherited disorders that predispose to RCC. These include:

234

von Hippel-Lindau Syndrome Dominantly inherited, inactivation of the von Hippel-Landau tumor-suppressor gene, associated with multi-systemic tumors, and a cumulative risk of RCC of 70% by the age of 60. Renal cell carcinoma is the most common cause of death among persons with this syndrome [1].
Hereditary Papillary Renal Carcinoma Autosomal dominant, associated with multifocal bilateral papillary RCC [1]. Tuberous Sclerosis Complex Autosomal dominant, manifestations include seizures, mental retardation, and hamartomatous lesions of many organs. The kidneys are affected by angiomyolipomas and cysts, and more rarely by RCC [1].

Environmental Factors
Cigarette Smoking Strongest environmental associations. The attributable risk of cigarette smoking for RCC is estimated to be between 21 and 30% among men and 9 and 24% among women [1, 2]. Cigarette smoking is a definitive risk factor for RCC and a dose-response relationship has been demonstrated [3].

Obesity, Diet and Physical Activity Consistent data (mainly case-control) have shown an association between obesity and RCC, especially among women, although this is an important risk factor in both genders. It has been postulated that obesity-related sex hormone stimulates renal cell proliferation and growth. The attributable risk of obesity for RCC is estimated at 13% to 25% [1]. Consuming fruits and vegetables appears to be protective, especially dark green and cruciferous vegetables [1, 4]. In a case-controlled study, high consumption of processed meat (hamburgers and sausage) was associated with an increased adjusted Odds Ratio (1.4, 95% CI = 1.1 1.8) and 1.5, 95% CI = 1.2 2.0) [4]. There are also some data suggesting that leisure time physical activity may be protective [1, 5]. Occupational Factors Exposures to asbestos and to high levels of the industrial solvent trichlorethylene have been associated with increased risk of RCC [1, 3]. In a case-control study [6] conducted in Iowa, U.S. of 406 incident cases of renal cell carcinoma and 2 434 referents, an elevated OR was reported for a number of occupations: among men these included mechanics and repairers (1.9, 95% CI = 1.2 - 2.9), farm product vendors (4.4, 95% CI = 1.3 15.5), guards (5.4, 95% CI = 1.4 - 20.7), among women employees of colleges and universities (7.6, 95% CI = 2.3 - 25.6).

235

Medical Conditions and Related Exposures

Radiation Therapy Women treated with RT for cervical cancer and men treated with RT for testicular cancer have a small, but significant increased risk of RCC. Other evidence of the importance of exposure to ionizing radiation is that patients with ankylosing spondylitis treated with radiation also have increased risk of RCC [1]. Renal Diseases There is strong evidence of an association between end-stage renal disease and RCC. There is also some evidence for an increased risk with nephrolithiasis [1]. Phenacetin abuse and analgesic nephropathy have been linked to risk for RCC [3].

Clinical Presentation About 50-60% of patients present with hematuria, other frequent manifestations are abdominal pain and mass in the abdomen or flank. Hypochromic anemia is often present. The classical triad: hematuria, abdominal pain and flank or abdominal mass occurs in 10 to 20% of patients. However, RCC may be clinically occult. Due to imaging techniques, the presentations have changed. Now, RCC is more frequently detected incidentally with abdominal imaging. Non-specific manifestations include: hypertension, weight loss, fever, night sweats. A number of paraneoplastic syndromes have been described in association with RCC. These include: erythrocytosis, hypercalcemia, hepatic dysfunction (Stauffers syndrome), acquired dysfibrinogenemia [1-3]. Differential Diagnosis The differential diagnosis of a renal mass includes: Renal cysts Benign solid neoplasm (adenoma, angiomyolipoma, oncocytoma) Inflammatory lesions (pyelonephritis, abscess) Metastatic disease (especially from melanoma). Classification and Grading According to the Robson classification system [2]:
Stage I = Confined to kidney Stage II = Extends through renal capsule, but confined to Gerotas fascia Stage III = Involves renal vein or vena cava (IIIA) or hilar lymph nodes (IIIB) Stage IV = Local invasion to adjacent organs or distant metastases.

236

Treatment and Prognosis

Radical or partial nephrectomy is the curative treatment. Among patients with apparently localized disease, approximately half will develop distant metastases. Overall 5-year survival rate is 40 to 50%, whereas at stage I-II (localized to kidney) 5-year survival is 80 to 95%. About 33% of patients are diagnosed with metastases, but the survival is variable with metastatic disease [1]. Renal cell carcinoma is refractory to chemotherapy. Biological therapies (interferon and interleukin-2) yield about 10-20% response, but these are durable in less than 5% of patients [2, 3]. Because of the toxicities and ineffectiveness of therapy and that up to 10% of patients with metastatic disease have long periods of stability, observation alone has been an option [2]. However, Linehan et al. [3] contend that this should not be considered a true option, since about 99% of these patients relapse and require therapy. Expression of proliferating cell nuclear antigen or MIB-1/Ki-67 antigens associated with cell proliferation is a poor prognostic factor [3].

14.2 Current approach to 1 diagnosis & staging

Ultrasound, CT and MRI are currently the methods used for primary diagnosis, as well as for staging of renal cell carcinoma. Intravenous pyelography, which was previously used in the work-up of renal masses, is neither sensitive nor specific, and has been completely replaced [3]. Linehan et al. [3] note: determining whether a space-occupying renal mass is benign or malignant can be difficult (p. 1366). Earlier studies have reported that the most frequent space-occupying renal lesions were benign cysts [3]. Such findings may be subject to re-evaluation as the diagnostics improve. Israel and Bosniak [7] point out that the rapid developments in CT and MRI technology have improved the detection and characterization of kidney masses, and predict that this should result in earlier diagnosis of renal cell carcinoma with better cure rates. A recent comparison [8] between CT and T1 as well as T2 weighted MRI for the diagnosis of 69 cystic renal masses in 59 patients, reveals that an increased wall or septal thickness was depicted by MRI in 10% of cases, leading to classification upgrade in 6 cases. In two lesions the enhancement characteristics of MRI and CT were different. Histologic findings in 25 of the lesions revealed that 20 were malignant and 5 were benign. While the findings on CT and MRI were generally similar, the additional findings on MRI lead to an upgraded classification, which

237

influenced patient management. MRI is also particularly helpful for evaluating involvement of the inferior vena cava [2].

14.3 Initial studies using in vivo MRS

There are very limited data available using in vivo MRS in the study of renal cell carcinoma and other pathological conditions of the kidney. Physiological motion artefact has been a major challenge. Katz-Brull et al. [9] addressed this problem by implementing multiple-breath holds to diminish the phase and frequency shifts as well as outer voxel contamination associated with abdominal and thoracic respiratory motion. Star-Lack et al. [10] employed a phase regularization procedure to improve SNR in the face of severe motion artefact (see Chapter 2). Kim et al. [11] applied MRI and MRS1 among 5 patients with biopsy-proven RCC at various stages. The authors [11] used a saddletype flexible surface coil. With STEAM the main resonances were sorbitol (3.85 ppm), trimethylamines2 (TMA) (3.25 ppm), and two peaks at 2.8 and 2.2 ppm, which were not identified. With PRESS the resonances at 3.25 ppm corresponding to TMA and at 1.35 ppm corresponding to lactate were seen in 1 patient with advanced stage RCC. The patient with low-grade tumor showed only TMA at 3.25 ppm. The authors note that these spectral patterns were different from those of the normal kidney. The TMA resonance may also rise with other, non-neoplastic diseases of the kidney. Mairiang et al. [12] performed MRS3 among 32 patients with renal stone disease (RSD) and 9 healthy volunteers. They report non-significantly increased peaks at 3.25 (corresponding to TMA), 3.6 and 3.9 ppm (both of the latter attributed to osmolytes which include glycerophosphocholine, betaine and myoinositol) among the patients with RSD. 31 P MRS has been applied in vivo among patients with renal allografts. A significant correlation has been found between the PME to Pi ratio and renal function [13]. There was also a correlation between the -ATP to Pi ratio and 3 year renal survival [14].

14.4 In vitro MRS studies of RCC

Investigation using in vitro MRS provides substantial insight into the metabolic features distinguishing renal cell carcinoma from normal

Kim et al. [11] used STEAM TR/TE=2200/20ms, and PRESS, TR/TE= 2200 / 288 ms. It should be noted that cholines are trimethylamines. Recall that the constituents of total choline resonate at 3.21 3.23 ppm. 3 E. Mairiang, et al. [12] used 1.5T, STEAM, single voxel, TR/TE=2000/15 ms. SAGE/IDL commercial software with 2 Hz line broadening.
2

238

kidney. Moreover, preliminary data suggest that various subtypes of RCC, which differ in aggressiveness, can be identified. Moka et al. [15] used HRMAS to compare normal renal tissue to RCC. They also used TOCSY and 2D inverse-detected 1H-13C heteronuclear correlation spectroscopy. The tumors were found to be mainly characterized by increased lipid content. The authors state: these are the first reported results on human tumor tissues using this technique and the approach offers potential for the rapid classification of different types of tumor tissue (p. 125). Sugars and amino acids were in the region from 3.4 to 4.1 ppm. Assignments for biopsy samples of normal renal cortex and RCC were as follows: _______________________________________________________
L1 = 0.7 ppm L2 0.9 ppm L3 1.3 ppm L4 1.6 ppm L5 2.05 ppm L6 2.25 ppm L7 2.8 ppm L8 4.05 ppm L9 4.3 ppm L10 5.2 ppm L11 5.3 ppm Axial C18 methyl of cholesteryl lipids CH3 (CH2) n Isoleucine, leucine and (maybe) valine (CH2) n Lactate CH2 CH2 CO CH2 CH= Glutamine CH2CO = Glutamine =CHCH2 CH= Aspartate amino acids and sugars CH2 of glyceryl CH of glyceryl =CH

________________________________________________________ Applying HRMAS proton MRS to examine 11 matched pairs of normal and tumor samples from the renal cortex, Tate et al. [16] used a computer-based pattern recognition technique to classify the tissues as normal versus tumor. Via principal component analysis, the spectral intensities showed a clear separation of the two classes. With linear discriminant analysis, it was possible to distinguish between normal and tumorous renal cortex (including one metastatic tumor from primary lung cancer) with 100% accuracy. The results are summarized in Table 14.1, with respect to informative metabolites. The authors [16] conclude: the spectra give rise to a plethora of sharp, wellresolved NMR peaks which are rich in biochemical information and variations in the levels of a number of these endogenous metabolites can be used to distinguish normal human kidney cortex tissue from renal cell carcinoma (p.70). The region between 3.00 and 4.00 ppm

239

normally contains carbohydrates and -CH resonances from amino acids, while the tumors showed increased intensity at the sites assigned to fatty acyl side chains of lipids. Table 14.1 The 20 most important correlations with NMR data point intensity for class (normal = 1, tumor = 2)
(From data of Tate et al. [16])

Chemical Shift (ppm)

4.00 3.96 3.93 3.90 3.84 3.58 3.55 3.52 3.48 3.37 3.17 3.10 2.98 2.66 2.61 2.31 1.66 1.55 1.31 1.27

Direction
Negative Negative Negative Negative Negative Negative Negative Negative Negative Negative Negative Negative Negative Negative Negative Negative Negative Positive Positive Positive

Possible Assignment
Phosphoethanolamine Myoinositol Aspartate Glucose Glucose Glycine Glycine Choline Glucose, Taurine Glycerophosphocholine Choline, Taurine Creatinine Dimethylglycine Aspartate Citrate Glutamate Arginine Lipid Lipid Lipid

240

Tosi et al. [17] examined tumor samples from 10 patients with RCC and compared these with normal surrounding healthy tissue. They note that renal osmolytes can be considered as markers of renal function, and that a decrease in these osmolytes is a hallmark of cancer. More recently, this group of authors [18] applied 13C MRS to lipids extracts from 3 human renal cancersthe presence of free cholesterol, high levels of unsaturated fatty acids, phosphatidylcholine, and a very high fatty acid to cholesterol ratio makes the lipid profile of a rare chromobophobe cell carcinoma very similar to that of an oncocytoma. In contrast, clear cell carcinomas showed nearly fully esterified cholesterol and much lower unsaturated fatty acids. The authors point out that since chromophobe cell carcinomas have a better prognosis than clear cell carcinomas, the finding of a lipid profile similar to benign kidney tumors could be important. They emphasize that cholesteryl esters and high unsaturated fatty acids might be a marker of poor (clear cell) versus good (chromophobic cell) prognosis, respectively. Yoshimitsu et al. [19] reported that clear cell RCC show a higher signal loss ratio (correlated with fat staining) compared to other RCC. Tugnoli et al. [20] also applied NMR and High Performance Liquid Chromatography to 13 samples of renal carcinoma, 9 from surrounding healthy parenchyma and 2 from healthy cortex and medulla. They confirmed that there is a difference in osmolyte concentration between the cortex and medulla, and that osmolyte concentration decreases markedly in renal cancer. In vitro 31P spectroscopy is also informative with respect to renal cell carcinoma. Applying 31P spectroscopy to blood samples from 5 patients with renal cell carcinoma, from patients with other cancers and from 15 healthy volunteers, Kuliszkiewicz-Janus et al. [21] found that among the patients with renal cell carcinoma and digestive cancers there was a decrease in lysophosphatidylcholine compared to the volunteers. Sullentrop et al. [22] applied 31P spectroscopy on blood samples from 29 patients with RCC prior to nephrectomy, 19 healthy volunteers, and 3 patients with other renal tumors (renal metastases, renal pelvic carcinoma and 1 benign renal tumor), as well as in 8 patients from the first group 6 months post-nephrectomy. Lysophosphatidylcholine and phosphatidylcholine concentrations were lower in patients with RCC compared to the volunteers. With remission, the low phospholipid concentrations normalized. The authors [22] state: the deviations in phospholipid concentrations observed may be attributable to systemic effects caused by the tumor as well as changes in enzyme activity (p. 60).

241

References
[1] P. Lindblad, H-O Adami, Kidney cancer, in: H-O. Adami, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, Oxford, 2002, p.467-485. [2] H.I. Scher, R.J. Motzer, Bladder and renal cell carcinomas, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 604-608. [3] W.M. Linehan, B. Zbar, S.E. Bates, M.J. Zelefsky, J.C. Yang, Cancer of the kidney and ureter in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 1362-1396. [4] J. Hu, Y. Mao, K. White, Canadian Cancer Registries Epidemiology Research Group, Diet and vitamin or mineral supplements and risk of renal cell carcinoma in Canada, Cancer Causes Control. 14, 705-714 (2003). [5] S. Mahabir, M.F. Leitzmann, P. Pietinen, D. Albanes, J. Virtamo, P.R. Taylor, Physical activity and renal cell cancer risk in a cohort of male smokers, Int. J. Cancer 108, 600-605 (2004). [6] Y. Zhang, K.P. Cantor, C.F. Lynch, T. Zheng, A population-based case-control study of occupation and real cell carcinoma risk in Iowa, J. Occup. Environ. Med. 46, 235-240 (2004). [7] G.M. Israel, M.A. Bosniak, Renal imaging for diagnosis and staging of renal cell carcinoma, Urol. Clin. N. Am. 30, 499-514 (2003). [8] G.M. Israel, N. Hindman, M.A. Bosniak, Evaluation of cystic renal masses: comparison of CT and MR imaging by using the Bosniak classification system, Radiology 231, 365-371 (2004). [9] R. Katz-Brull, N.M. Rofsky, R.E. Lenkinski, Breathold abdominal and thoracic proton MR spectroscopy at 3T, Magn. Reson. Med. 50, 461-467 (2003). [10] J. Star-Lack, E. Adalsteinsson, G.E. Gold, D.M. Ikeda, D.M. Spielman, Motion correction and lipid suppression for 1H magnetic resonance spectroscopy, Magn. Reson. Med. 43, 325-330 (2000). [11] D.Y. Kim, K.B. Kim, O.D. Kim, J.K. Kim, Localized in vivo proton spectroscopy of renal cell carcinoma in human kidney, J. Korean Med. Sci. 13, 49-53 (1998). [12] E. Mairiang, P. Hanpanich, P. Sriboonlue, Proton magnetic resonance spectroscopy of the kidney in renal stone disease, Magn. Reson. Imaging, 20, 777-779 (2002). [13] M.B. Niekisch, D. Von Elverfeldt, A.E. Saman, J. Hennig, G. Kirste, Improved pretransplant assessment of renal quality by means of phosphorus-31 magnetic resonance spectroscopy using chemical shift imaging, Transplantation, 77, 1041-1045 (2004). [14] K. Seto, H. Ikehira, T. Obata, et al., Long-term assessment of posttransplant renal prognosis with 31P magnetic resonance spectroscopy, Transplantation, 72, 627-630 (2002). [15] D. Moka, R. Vorreuther, H. Schicha, et al., Biochemical classification of kidney carcinoma biopsy samples using magic-angle spinning 1H nuclear magnetic resonance spectroscopy, J. Pharm. Biomed. Anal. 17, 125-132 (1998). [16] A.R. Tate, P.J.D. Foxall, E. Holmes, et al. Distinction between normal and renal cell carcinoma kidney cortical biopsy samples using pattern recognition of 1H magic angle spinning (MAS) NMR spectra, NMR Biomed. 13, 64-71 (2000). [17] M.R. Tosi, V. Tugnoli, G. Bottura, et al., In vitro MRS and HPLC studies in human renal cell carcinomas, Oncol. Rep. 7, 1355-1358 (2000). [18] M.R. Tosi, A. Reggiani, V. Tugnoli, Are molecular features of a chromophobic cell renal cell carcinoma correlated with clinical findings? Int. J. Molcular Med. 12, 99-102 (2003).

242

[19] K. Yoshimitsu, H. Honda, T. Kuroiwa, et al., MR detection of cytoplasmic fat in clear cell renal cell carcinoma utilizing chemical shift gradient-echo imaging, J. Magn. Reson. Imaging 9, 579-585 (1999). [20] V. Tugnoli, A. Reggiani, R. Beghelli, V. Tomaselli, A. Trinchero, M.R. Tosi, Magnetic resonance spectroscopy and high performance liquid chromatography of neoplastic human renal tissues, Anticancer Res. 23, 1541-1548 (2003). [21] M. Kuliszkiewicz-Janus, W. Janus, S. Baczynski, Application of 31P NMR spectroscopy in clinical analysis of changes of serum phospholipids in leukemia, lymphoma and some other nonhaematological cancers, Anticancer Res. 16, 1587-1594 (1996). [22] F. Sullentrop, D. Moka, S. Neubauer, et al., 31P NMR spectroscopy of blood plasma: determination and quantification of phospholipid classes in patients with renal cell carcinoma, NMR Biomed. 15, 60-68 (2002).

243

Chapter 15

Hepatic, Gastrointestinal and other Tumors

_______________________________________________________________________________

This chapter provides a review of a number of hepatic and gastrointestinal (GI) malignancies for which MRI has been of major diagnostic importance, but for which, as yet, there is limited experience with MRS and MRSI. The available MR spectroscopic data, much of which are in vitro, will be presented for these liver, GI and other cancers. We will also describe recent innovations in MR imaging that have improved detection of these malignancies. The role of advanced CT-based diagnostic modalities will be discussed, together with PET, illustrating the value of molecular imaging. .

15.1 Functional Anatomical Imaging using PET and advanced CT-based methods
It is well known that tumor cells frequently rely upon energy generated from glycolysis to fuel their requirements for rapid replication. This is because cancerous degeneration of tumor cells is associated with loss of the more efficient mode of aerobic ATP production via the Krebs cycle. As a result, cancer cells show increased glucose consumption for a given level of energy production. Glycolysis also increases because of the activation of the hexose monophosphate pathway, which provides the carbon skeleton for deoxyribonucleic acid and ribonucleic acid synthesis in growing tumors [1]. With the emergence of PET [2] these insights into tumor cell metabolism could be used to develop biological whole-body imaging of cancer via the glucose analogue 18F deoxyglucose. After injection, FDG is taken up by tumor cells and phosphorylated by hexokinase to FDG-6-PO4. The rate of FDG uptake is proportional to the rate of glycolysis. However, unlike glucose-6-phosphate, FDG-6-PO4 is not

244

further metabolized via the glycolytic pathway. Instead, it becomes trapped inside the cells, since tumor cells do not contain enough of the enzyme glucose-6-phosphatase that would reverse this reaction during the time needed for imaging. As a consequence, the amount of trapped FDG-6-PO4 registers glycolysis that occurs in cells throughout the body, and this is imaged at 45 minutes to 1 hour after FDG has been injected [1]. As mentioned earlier in this book, PET has a number of limitations with respect to cancer diagnostics. Some of these include: Spatial resolution is about an order of magnitude less than for MRI [3], False negative results for tumors with low metabolic activity, False positive results for inflammation / infection, Reliance upon single markers that are not entirely sensitive or specific for malignant activity.

Notwithstanding these limitations, we have noted that PET-CT is currently the most widely applied modality for functional-anatomical imaging in oncology. It has shown superior sensitivity and specificity with respect to anatomical techniques alone for a number of cancers [1, 3]. Moreover, a host of new radiopharmaceuticals have been and are being developed in positron-emitting forms, and these represent promising lead molecules for oncologic imaging (p. 683) [3]. This new area is beyond the scope of this book and the reader is referred to the rapidly expanding literature on this topic. We will briefly summarize the diagnostic efficacy of PET for hepatic, GI and some other tumors, not previously presented, for which this modality has been successfully applied. 15.1.1 FDG-PET for cancerous involvement of the liver

Metastases to the liver In their meta-analysis, comparing ultra-sound, CT, MRI and PET, Kinkel et al. [4] concluded that for a given level of specificity, FDGPET is the most sensitive non-invasive imaging modality for the diagnosis of hepatic metastases. The overall sensitivity of PET was reported to be 0.90 (95% CI = 0.82 0.96) [4].

245

Hepatocellular Carcinoma With respect to hepatocellular carcinoma (HCC), since the metabolic properties of normal liver tissue are retained in part,1 FDG-PET uptake in HCC is often comparable to that of normal liver. As a consequence, FDG-PET has low sensitivity for detecting HCC and would not appear to be a useful modality for routine staging of HCC [5]. 15.1.2 FDG-PET and virtual CT endoscopy in GI cancers

Colorectal Cancer With respect to the initial diagnosis of colorectal cancer, FDG-PET has a low specificity. This is because of the large numbers of false positive findings due to inflammation.
Dilemmas regarding approach to screening for colorectal cancer Currently, consensus is lacking as to the best approach to screen for colorectal cancer, which is among the most common cancers in western populations [6]. Tests for occult fecal blood have about a 50% false negative rate; screening recto-sigmoidoscopy is becoming less efficacious as the proportion of cancers arising more proximally is increasing, whereas routine colonoscopy would be logistically difficult. Analysis of stool for specific ras protooncogene mutations is being investigated for possible screening purposes [7]. Screening of high risk groups for colorectal cancer Screening of groups at high risk has been generally recommended. These groups include those with familial adenomatous polyposis, hereditary nonpolyposis colon cancer, first-degree relations who have had colorectal cancer or adenomatous polyps, personal history of adenomatous polyps or colorectal cancer or inflammatory bowel disease. Screening for early detection has shown a survival benefit, since early stage colorectal cancer has a 90% 5year survival rate [8]. However, as emphasized by Saar et al. [9], these risk factors comprise only a minority of those who develop colorectal cancer, which is the second leading cause of death in western industrial countries. FDG-PET for staging and restaging of colorectal cancer For staging and restaging of colorectal cancer, FDG-PET shows superior sensitivity and specificity compared to CT, according to the reviews by Czernin and Phelps [1] and by Pomper [3]. However, Tzimas et al. [10] present a different opinion in their review, stating: currently, there is no role in standard clinical practice for PET scan evaluation in screening, primary
1

High levels of glucose-6-phosphatase, resulting in low tumor retention of FDG.

246

detection, evaluation of local or lymph node invasion or evaluation of extracolonic primary metastatic disease, because its overall performance is not better than what is currently used the main role for PET scan at this moment is in identifying extrahepatic disease in order to assess resectability in cases of recurrence and hepatic metastases (p. 649). Virtual CT colonoscopy Virtual colonoscopy or CT colonography, which generates high-resolution, 3D images, has been proposed for primary detection, as well as for staging of colorectal cancer; the radiation dose is about 20% lower than the usual dose for double-contrast barium enema [11, 12]. A recent multi-center study [13] of CT colonography, that includes 615 participants with a total of 827 lesions, reveals that for lesions sized 6 mm, the sensitivity was 0.39 (95% CI = 0.296 0.484) and for lesions 10 mm the sensitivity was 0.55 (95% CI = 0.399 0.70). These results are significantly lower than for conventional colonoscopy with sensitivities approaching 100%. The authors [13] conclude that this is an exciting new technology (p. 1718) but that computed tomographic colonography by these methods is not yet ready for widespread clinical application. The technique and training need to be improved (p. 1713).

Other GI cancers
Gastric carcinomas There is limited data on the use of FDG-PET for gastric carcinomas. Karpeh et al. [14] describe a pilot study from Memorial Sloan-Kettering Cancer Center to evaluate the feasibility of FDG-PET imaging for pre-treatment staging. Preliminary results indicated that FDG-PET had a sensitivity of 60%, a specificity of 100% and overall accuracy of 94% for identifying gastric carcinoma. The most consistent findings were for FDG uptake in the primary lesions, suggesting that FDG-PET might be helpful in assessing response to therapy, as well as for staging. Using CT alone for pre-treatment staging, up to 50% of patients are reported to have more extensive disease at laparotomy than was predicted [14]. Virtual CT gastroscopy Preliminary results suggest that 3D CT used in combination with virtual CT endoscopy can aid in the identification of gastric lesions, and that virtual CT gastroscopy can identify both intraluminal and submucosal components of gastric lesions. Virtual gastroscopy, however, still has limited spatial resolution and does not depict flat or small gastric lesions, nor does it permit detailed assessment of the mucosa. As stated by Oto [12]: further evaluation of this new technique is warranted in the diagnosis and staging of gastric cancers (p. 238). Approach to screening of high-risk populations In high-risk regions, such as Japan, mass screening programs using doublecontrast barium radiographs or upper endoscopy have a reported sensitivity and specificity of approximately 0.90. This has yielded very important clinical

247

results; in that up to 40% of the newly diagnosed patients have early gastric carcinomas, with a high cure rate [14]. Esophageal cancer For esophageal cancers FDG-PET may have a role in the detection of Stage IV disease [5]. There are also initial reports of virtual CT endoscopy for the evaluation of the esophagus [15, 16]. Using a multidetector CT after esophageal distension, Mazzeo et al. [16] assessed 33 patients who had various esophageal pathologies, including 12 carcinomas. Spiral CT yielded a sensitivity of 0.84 and specificity of 0.87 for the detection of pathologic wall thickening. The authors conclude: evaluation of the esophagus with multidetector CT is a promising technique and easy to use, allowing panoramic exploration, virtual endoluminal visualization, accurate longitudinal and axial evaluations, and simultaneous evaluation of T and N parameters (p. 2). Pancreatic cancer A somewhat improved sensitivity 0.88 (range 0.85 0.92) is gained with PET for pancreatic cancer compared to CT, whose negative predictive value is 0.74 (range 0.71 - 0.76). The specificity of PET for pancreatic cancer is reported to be 0.82 (range 0.77 0.85) [3]. Kalra et al. [17] emphasize the complementarity of PET to CT for patients with suspected pancreatic cancer. PET is particularly helpful for detecting distant metastases and for evaluation of suspected recurrence, where CT may be equivocal due to anatomical distortion after therapy. Moreover, the absence of FDG uptake at 1 month post-chemotherapy may be indicative of improved prognosis [18]. Multi-slice 3D spiral CT cholangiography has been applied for the clinical diagnosis of pancreatic and biliary diseases including carcinoma of the head of the pancreas and periampullar carcinoma [19]. Spiral CT has a reported positive predictive value of 1.0, negative predictive value of 0.56 and overall accuracy of 0.70 for unresectable pancreatic carcinoma [17].

15.1.3

FDG-PET for diagnosis of some other cancers

FDG-PET for detection and staging of lung cancer For characterizing solitary pulmonary nodules, FDG-PET provides 0.95 sensitivity (95% CI = 0.93 0.97) with a specificity of 0.83 (95% CI = 0.80 0.86). PET has been shown to be superior to CT for staging of the mediastinum: with a sensitivity of 0.89 (95% CI = 0.86 - 0.91) and specificity of 0.94 (95% CI = 0.91 0.96) compared to a sensitivity and specificity of 0.67 and 0.73, respectively for CT [1]. The improved staging of lung cancer provided by whole-body PET, has helped to avoid unnecessary thoracotomies, and has impacted upon management of an estimated 41% of patients with lung cancer [3].

248

FDG-PET for staging of melanoma Whole-body PET imaging has become an important part of staging in patients with melanoma, where its diagnostic accuracy is very high (sensitivity 0.92 (95% CI = 0.88 0.96)) and specificity 0.90 (95% CI = 0.83 0.96) [1, 3].

15.2 MRI and MRS in hepatic cancers

15.2.1 Hepatocellular carcinoma MRI findings in hepatocellular carcinoma In their review, Annovazzi et al. [5] state that currently the most accurate diagnostic modality for hepatocellular carcinoma is contrast MRI, with sensitivity ranging from 0.82 to 0.96. With T2 weighting, MRI shows certain characteristic features in HCC; these include mosaic or nodules-in nodule patterns, encapsulation and satellite lesions [20, 21]. After injection of contrast HCC also typically shows dynamic signal enhancement, distinct from surrounding liver. This is due to the prominent arterial blood supply of HCC. MRI is also helpful for detection of early HCC arising inside regenerating nodules associated with cirrhosis. In such cases, -fetoprotein is often normal, and biopsy may even be negative [20]. Yu et al. [22] suggest that the combination of superparamagnetic iron oxide (see Subsection 15.2.3 on metastases) and gadolinium would further improve detection of hepatocellular carcinoma, distinguishing HCC from regenerative or dysplastic nodules and from focal nodular hyperplasia. In vitro MRS studies of hepatocellular carcinoma Thus far, primarily in vitro MRS studies have been performed specifically with respect to HCC (see later Subsections 15.2.3 and 15.2.4 for in vivo MRS studies comparing HCC with liver metastases, and on response to therapy in HCC). Using 1D as well as 2D COSY 1H MRS, Soper et al. [23] compared liver specimens that were classified as normal (n = 31), cirrhotic (n = 59) and as hepatocellular carcinoma (n = 32). The HCC specimens were consistently distinguished from normal and cirrhotic liver by reduced lipid and carbohydrate residues, and by increased choline-containing compounds. The ratio of choline to creatine (3.2:3.0 ppm) was greater than 4 for all the HCC samples and below 3 for all the cirrhotic and normal specimens. Applying a statistical classification strategy to the MR spectra, these authors [23] achieved 100% accuracy

249

in distinguishing HCC from normal hepatic tissue. There were 2 misclassifications when the cirrhotic and HCC samples were assessed, and 8 of the samples were non-reliably classified (fuzzy). The distinctions between normal and cirrhotic liver were poorer, with 5 misclassifications and 16 fuzzy classifications. The authors [23] note that MRS provides helpful information for in vitro diagnosis of HCC in small biopsies, but that in vivo implementation is still difficult due to respiratory motion and frequently small lesion size. However, if successful, the authors underscore that in vivo MRS would offer the potential to non-invasively diagnose HCC prior to surgical intervention.
In vitro MRS studies of bile in hepatocellular carcinoma Patients with hepatocellular carcinoma, as well as other hepatobiliary malignancies, have been found to show elevated lactate levels in bile compared to healthy individuals or those with non-malignant hepatic and/or biliary disease. These findings were obtained by applying in vitro 1H MRS to bile samples [24].

15.2.2 Hepatic Lymphomas As noted in Chapter 12, a study by Dixon [25] revealed that among 11 patients with Hodgkins and non-Hodgkins lymphoma involving the liver, 6 showed elevated PME/inorganic phosphate ratios on 31P MRS of the liver. In 2 of these patients this ratio dropped to within the normal range with clinical remission, whereas four patients in whom this ratio remained high after chemotherapy died of progressive disease. 15.2.3 Liver Metastases MRI findings for metastatic hepatic lesions On MRI metastatic hepatic lesions are generally slightly hypo-intense on T1 weighted images and hyper-intense with T2 weighting. Central necrosis usually shows a different intensity compared to surrounding tissue. Contrast agents containing superparamagnetic iron oxide (SPIO) particles have improved the accuracy by which liver metastases can be detected via MRI. SPIO acts as a negative contrast agent in T2 weighted images, and is taken up by Kuppfer cells. Metastatic hepatic lesions do not contain Kuppfer cells and, therefore, do not take up these SPIO particles. Consequently, the metastases show relatively high intensity compared to the reduced signal from the surrounding normal hepatic tissue [20, 26].

250

In vivo 31P MRS In vivo 31P MRS has been used to evaluate hepatic metastases in a few investigations. Brinkman et al. [27] applied in vivo 31P MRS in 24 patients with liver metastases and in 20 volunteers. Focal hepatic tumors were reportedly detectable with a surface coil at a distance within the coil radius. At the defined VOI, phosphomonoester to ATP and phosphodiester to -ATP ratios were significantly elevated in the patient group. The authors [27] concluded: detection of small tumor volumes within a VOI filled by less than 50% by tumor is possible, with results statistically different from that in normal volunteers (p. 56). 15.2.4. MRS Studies on Heterogeneous Liver Cancers In vivo 31P MRS and MRSI In an early study by Francis et al. [28] applying in vivo 31P MRS, patients with liver metastases from a variety of primary cancers and those with hepatocellular carcinomas did not differ in spectral characteristics or metabolite ratios. Among the patients in whom tumors occupied over 50% of the analyzed section, phosphomonoester to -ATP ratios were significantly elevated compared to normal. In another clinical early study of the liver, Cox et al. [29] applied in vivo 31P MRSI among 28 healthy adults and 32 patients with a variety of hepatic malignancies. As shown in Table 15.1, although there was some overlap, the mean PME to PDE ratio was significantly higher among patients. In vitro analysis was also performed in 6 liver samples that were histologically normal and in 7 containing tumors of various histologies. All of the latter showed an increase in phosphoethanolamine and phosphorylcholine, and a decrease in glycerophosphorylcholine and glycerophosphorylethanolamine. The in vivo and in vitro findings were compared in 5 of the patients, indicating that the increased PME was due to an increase in phosphoethanolamine and phosphorylcholine and the decreased PDE to a decrease glycerophosphorylcholine and glycerophosphorylethanolamine.

251

Table 15.1 Phosphomonoester to phosphodiester ratio in patients with various hepatic malignances and in healthy volunteers
(From data of Cox et al. [29]) 32 Patients with Hepatic Malignancies
Mean (range)

28 Healthy Adults
Mean (range)

Phosphomonoester to Phosphodiester Ratio

0.68 (0.15 2.38)

0.23 (0.15 0.41)

denotes p < 0.001

Assessment of response to embolization with 31 P-MRS Albeit from small clinical studies and with some contradictory results, there are data suggesting that 31P MRS may provide useful insights concerning response to therapy among patients with hepatic involvement from metastatic and primary malignant processes. Ljungberg et al. [30] found among patients with liver metastases from neuroendocrine tumors, who were treated with hepatic artery embolization, that the PME/Pi ratios were significantly associated with response to therapy. An earlier study by Meyerhoff et al. [31] using in vivo 31P MRS, included 5 patients with primary and metastatic liver cancers treated with hepatic embolization combined with intra-arterial administration of cytostatics drugs (chemoembolization). As an acute response to chemoembolization, phosphomonoester, phosphodiester and ATP decreased, while Pi either stayed the same or increased. Long-term follow-up revealed decreased phosphomonoester to ATP ratios, and increased ATP concentrations. There were no morphologic changes seen on standard MRI or CT. Different findings were reported that same year from a study by Schilling et al. [32], where in vivo 31P MRS was applied among 10 patients with hepatic metastases from colorectal cancer and in 2 patients with hepatocellular carcinoma prior to and during chemoembolization. These authors reported a decrease in and -NTP portions of the spectra, as well as a rise in Pi. Subsequently, there was a marked increase in phosphomonoester, and a slight decrease in phosphodiester. They conclude: the effects of successful chemoembolization or local chemotherapy become apparent in the 31P MR spectrum during the first few hours after the start of therapy (p. 887).

252

Among 6 patients with hepatocellular carcinoma, those in whom arterial embolization was effective showed a decrease in ratio of ATP to Pi in a study by Taniguchi et al. [33].

In vitro 31P and 1H MRS studies of hepatic malignancies An in vitro study by Bell et al. [34] applying both 31P and 1H-MRS revealed that in addition to elevation in phosphoethanolamine and phosphocholine, liver malignancies showed high taurine, citrate, alanine, lactate and glycine. Besides reduction in glycerophosphorylcholine and glycerophosphorylethanolamine, creatine and threonine were also decreased in the hepatic neoplasms.

15.3 MR diagnostics in colorectal cancer

MRI is currently considered to be most useful for diagnosing colorectal cancer in the rectosigmoid region, since image quality is only minimally degraded by respiration and peristalsis. Endoluminal or phased array coils are used to provide very high spatial resolution [11]. While MRI offers better soft tissue contrast than CT, enabling the identification of the bowel wall layers, high quality MR imaging of the colon has not yet been achieved in clinical practice [35]. MRI has, however, become an important modality for detecting local extension of rectal tumors, and, as noted above, application of new contrast agents such as SPIO has improved the diagnostic yield of MRI with respect to liver metastases [20, 26]. 15.3.1 Virtual MR colonography Virtual colonography using MR acquisition is being developed. This technique is based upon the principles of contrast-enhanced 3D MR angiography and requires breath-hold. Three-dimensional data acquisition for MR-based colonography requires less than one minute, using 3D gradient-echo sequences [9]. Luboldt et al. [36] performed MR colonography (MRC) in 132 patients referred for colonoscopy because of suspicion of a colonic mass. Using conventional colonoscopy as the standard of reference, 13 of 14 lesions >10 mm in size were detected; the specificity was 99%. However, for smaller lesions the diagnostic accuracy was much poorer. Notably, only 18 of 39 lesions < 5 mm were detected. Papparlado et al. [37] carried out a similar study among 70 patients, among whom all but 5 had endoluminal colonic masses. MRC performed comparably to conventional colonography for detecting endoluminal masses 6mm. MRC failed to detect 4 of 6 adenomatous polyps 5 mm in diameter, but identified a synchronous cancer in the right colon of a patient with a

253

stenotic sigmoid carcinoma that was not traversed endoscopically. In a subsequent paper, Luboldt and Morrin [38] note that there are fewer data available concerning MR colonography compared to CT colonography for detection of colorectal masses. The potential advantages of MRI in avoiding exposure to ionizing radiation, plus the capability of multi-planar imaging, superior soft-tissue contrast and higher sensitivity to IV contrast are underscored by Oto [12], Saar et al. [9], and Luboldt and Morrin [38]. The disadvantages of MRC include lower spatial resolution, higher susceptibility to motion artefacts, as well as time and availability considerations [38]. Overall, virtual MRC would be an acceptable alternative to patients for screening compared to the currently available diagnostic tests such as double barium enema, colonoscopy or rectosigmoidoscopy, which either are associated with substantial discomfort, require sedation or entail exposure to ionizing radiation. Luboldt et al. [36] conclude: the performance of MR colonography, as currently implemented, for the detection of masses that exceed 10 mm in diameter warrants further consideration of this technique as a potent option in the diagnostic arsenal for colorectal mass screening (p. 388). The need for further optimization of this promising technique is underscored [38]. Ransohoff [39] emphasizes that at the present time, it is necessary to be cautious not to implement virtual colonoscopy too widely without sufficient regard for current technological problems that affect sensitivity. The level of sensitivity and specificity that virtual colonoscopy can achieve, in some specialized situations, is known. Yet the differences between what virtual colonoscopy can do and what it will do if applied in ordinary practice are so great that physicians must be cautious. There are many important steps yet to be taken in learning how to implement this new technology appropriately (p. 1774). 15.3.2 In vitro MRS findings in colorectal cancer While in vivo MRS studies of colorectal cancer are still lacking, there have been several investigations of colon cancer biopsy specimens. T1 and T2 relaxation times In vitro results have shown that cancerous colon exhibits marked changes in T1 and T2 relaxation times, compared to healthy large bowel [40, 41]. However, the differences in measured relaxation times were not of sufficient magnitude to rely upon in vitro MR data alone to diagnose malignancy [41]. Mountford et al. [42] reported that neoplastic colon specimens associated with metastasis had T2 relaxation times > 350 ms; overall this was the case in 29 of 31 malignant colon specimens that they analyzed in vitro using proton MRS.

254

In vitro 1H MRS studies in colorectal cancer More recently, Mountford et al. [43] applied in vitro proton MRS and MRSI to compare 45 malignant colorectum samples with 43 normal specimens. The neoplastic samples showed resonances at 3.2 ppm from the N-trimethyl group of choline and at 2.2 ppm attributed to the acetyl group of sialic acid and CH2-CH2- of glutamine and glutamate. Using MRSI they identified the site of origin of MR-visible lipid as arising from the submucosa in normal specimens versus from the malignant cells in the cancerous colorectum. Thus, the spatial distribution of lipid was informative. The authors note: infiltrating carcinoma destroys normal architecture and cellular function, and lipid CSI reflects this disruption showing no consistent pattern in the distribution or intensity of the lipid {peak} (p. 1529). An in vitro study by Moreno et al. [44] using proton MRS at 9.4T compared 23 pairs of normal colonic mucosa with adenocarcinoma of the colon. They found a partial overlap between normal and adenocarcinoma in the ratio of the area of 3.2 ppm (trimethylamine-containing compounds) to that of 0.9 ppm (methyl of fatty acids). They also pointed out the possibility of artifactual decrease in that ratio due to high triglyceride content of the normal colonic submucosa. On the other hand, these authors reported that the taurine (3.4 ppm) to creatine (3.0 ppm) ratio produced excellent discrimination between normal mucosa and tumors groups (p. 111).
In vitro 2D COSY 1H MRS in colon cancer In a subsequent study, Moreno and Ars [45] compared 16 colon tumors to 10 normal mucosa biopsies applying 2D COSY proton MRS also at 9.4T. They found a significant increase in lactate, glutamate, aspartate, taurine, spermine, glutathione and glycerophosphoethanolamine in the tumors, with a significant decrease in myoinositol and scylloinositol in the cancers. These authors underscore the advantages provided by 2D COSY for detecting metabolites such as myoinositol, which were reportedly difficult to observe with 1D in vitro MRS.

In vitro 31P MRS studies in colorectal cancer Comparing malignant and normal human colon specimens using 31P MRS, Kasimos et al. [46] found that the cancers showed elevations in phosphomonoesters, glycerolphosphodiesters, and in the ratios of phosphorylethanolamine to phosphorylcholine, and of phosphomonoesters to inorganic orthophosphate. In addition, the neoplastic colon showed significantly decreased ratios of phosphocreatine to inorganic orthophosphate and to ATP. The authors considered these results as indicating that in the malignant colon cancer

255

specimens there were significant alterations in high energy metabolism, low energy metabolism and membrane metabolism (p. 527). Significantly elevated levels of lysophosphatidylcholine and phosphatidylcholine plasmalogen, and decreased sphingomyelin and phosphatidylethanolamine plasmalogen were found in human colon cancers compared to non-malignant specimens in a subsequent study by the same group of authors [47]. More recently, Merchant et al. [48] applied in vitro 31P MRS to examine colon cancer at various stages and grades. Elevated glycerol phosphate was found in higher stage or less well-differentiated tumors. Significantly increased sphingomyelin, -glycerol phosphate and glycerol 3-phosphoserine and lower phosphatidylinositol were observed in the lymph node positive samples. The authors of Ref. [48] suggest: these differences in the phospholipid and intermediate phosphate metabolite profiles identified through magnetic resonance spectroscopic and histopathologic analysis may provide important information regarding the nature of tumor and cell membrane metabolism. Differences in these profiles may identify markers useful for biologic behaviour, provide prognostic information and characterize the impact of the pathologic features of colon cancer (p. 1715). MRS studies in colorectal cancer in animals and in human cell lines There have been numerous studies of colorectal cancer in animals and in human cell lines using in vitro MRS. Proton MRS has been used to distinguish highly tumorigenic, human malignant colorectal cell lines from those with low tumorigenicity or from adenoma cell lines. Further discussion of this area is outside the scope of this book. The reader is referred to [49, 50], inter alia for more on this topic.

15.4 MR diagnostics in gastric cancer

15.4.1 Virtual MR gastroscopy While endoscopic sonography has been the conventional method for tumor staging, and CT for N-staging, recent advances in MRI technology have yielded results that are considered at least comparable to that of helical CT for staging gastric carcinoma [51, 52]. The basic approach has been to apply short-duration T1 weighted spoiled gradient echo sequences (see glossary) before and after administration of gadolinium. This technique has been effective for evaluating the extent of gastric wall invasion and peritoneal dissemination. Overall, MRI seems to be effective for assessing local and distant extents of gastric

256

carcinoma, but less well in the presence of coexistent gastritis or for small tumors present in a non-distended stomach [51]. There are also a few reports of virtual MR gastroscopy using high field magnets or other dynamic MRI studies to demonstrate the gastric wall layers. Motohara et al. [51] emphasize that the results obtained thus far suggest that MRI could eventually be applied for staging smaller cancers, that the need for improvement is greatest for N-staging, and that advances such as parallel imaging will render multiplanar, fast, and high-resolution MRI more feasible for evaluating gastric cancer in the near future(p. 382). 15.4.2 In vitro MRS findings in gastric carcinoma There are very limited in vitro, and as yet no in vivo, data on human gastric carcinoma. Using high field 1H MRS, Mun et al. [53] assessed 35 gastric specimens resected during surgery from patients with carcinomas of the stomach. Cancerous and non-cancerous tissues were compared. The malignant gastric specimens showed decreased lipid peaks, and increased choline and lactate. On the basis of their results, the authors suggest that high resolution in vivo proton MRS could potentially yield useful information for the clinical diagnosis of gastric carcinoma.

15.5 MR diagnostics in esophageal cancer

15.5.1 MRI and MR endoscopy of the esophagus As is the case for gastric carcinoma, endoscopic ultrasonography is the standard method that currently is the most accurate diagnostic modality for loco-regional staging esophageal cancer, yielding 61-92% accuracy for T stage and 75 to 85% for N-stage. However, endoscopic sonography has low soft tissue contrast between the cancer and the muscularis propria, whereas MRI provides better soft tissue contrast enhancement [54, 55]. MRI with T1 weighted sagittal views is helpful for assessing the extent and location of esophageal cancers. Tumor invasion of the periesophageal fat, involvement of regional lymph, mediastinum and adjacent structures can also be evaluated with MRI [55]. A pilot study by Dave et al. [54] reveals that MR endoscopy yielded accurate T-staging in 6 of 7 patients and N-staging in 5 of 6 patients with esophageal cancer who subsequently underwent esophagectomy.

257

MR and ultrasound concurred in 7 of 8 patients for T-stage and in 5 of 8 patients for N stage. The authors conclude: endoscopic MR imaging is safe and probably comparable to endoscopic {ultrasound}, but with a tendency to overstage the disease (p. 281). 15.5.2 In vitro MRS findings on esophageal cancer In Vitro 1H MRS A recent in vitro 1H MRS study [56] of 72 esophageal biopsies reveals that a statistical classification analysis distinguished normal esophageal tissue (n = 29) from adenocarcinoma (n = 35) and from Barretts epithelium (n = 41) with 100% accuracy. Barretts epithelium and cancers were distinguished with 98.6% accuracy (1 misclassification). The distinction between normal and cancerous esophagus used the spectral regions at 3.49 3.57 ppm, 3.23 3.28 ppm and 0.92 to 1.12 ppm. To differentiate Barretts epithelium from carcinoma, the regions used were at 3.05 3.13 ppm, 2.51- 2.57, 1.45 1.49 ppm, and 1.211.29 ppm. Different spectral categories of Barretts epithelium were also identified; these may help predict future risk of malignant transformation. In Vitro 31P MRS Merchant et al. in two studies [57, 58] examined membrane phospholipids in esophageal cancer using 31P MRS. They were able to quantify 18 individual phospholipids and found a number of correlations between percentage concentrations of various components, e.g. dimethylphosphatidylethanolamine, lysoethanolamine plasmalogen inter alia, with T-stage, nuclear grade, or presence of lymphatic invasion. Phosphatidyl glycerol significantly decreased with increasing tumor grade, and with lymphatic vessel invasion. The authors suggest: the decreasing phospholipid concentration correlates to the transformation of the cells from a less aggressive to a more aggressive phenotype (p. 187) [58]. Comparing normal esophageal tissue with cancerous lesions, they found significant differences in phosphatidylserine, sphingomyelin, phosphatidylinositol and phosphatidylcholine [57]. The overall line of research that these authors are pursuing is to provide a method to predict the biologic behavior of these tumors. This may help select patients who would benefit from specific therapies such as neoadjuvant and post-operative therapies (p. 187) [58].

258

15.6 MR diagnostics in pancreatic cancer

15.6.1 MRI and MR endoscopy of the pancreas According to Kalra et al. [17]: state-of-the-art MRI of pancreatic neoplasms is optimally performed with 1.5T gradient systems using phased-array torso coils to improve the signal-to-noise ratio, optimized with thin slice profiles and small fields of view Breath-hold acquisitions are obtained with fast spin echo (FSE) or gradient echo (GRE) sequences and echo planar imaging. A moderately T2 weighted FSE and single shot fast spin echo (SSFSE) should be obtained, followed by T1 weighted in-phase GRE and T1 weighted opposed-phase GRE. To evaluate cystic lesions of the pancreas, coronal and axial magnetic resonance cholangiopancreatography with SSFSE are usually obtained (p. 857). Virtual MR endoscopy was applied to the pancreas (virtual pancreatoscopy) in 26 patients with pancreatic cancer studied by Tanizawa et al. [59]. Clear virtual images were obtained in 20 of the patients, accessing cystic lesions as well as imaging the pancreatic duct behind a stricture. This procedure caused far less discomfort compared to actual endoscopic evaluation of the pancreas. T1 weighted spin echo MRI with and without fat suppression and immediate post-gadolinium spoiled gradient echo sequences are considered better than spiral CT imaging for the detection of small pancreatic cancers. Since pancreatic adenocarcinomas typically contain dense fibrotic tissue, they usually appear as slightly hypo-intense on T2 weighted images. They are often difficult to visualize unless they are necrotic. As fairly hypovascular neoplasms, they do not typically enhance, and appear hypo-intense compared to the surrounding normal pancreatic tissue [17]. T1 weighted spin-echo MRI may be better than dynamic contrast-enhanced CT imaging for determining vascular encasement. As noted in Chapter 12, pancreatic lymphomas are generally iso-intense to normal pancreas on both T1 and T2 weighted images. 15.6.2 MRS studies of pancreatic cancer in animals and in human cell lines As discussed earlier in this chapter, molecular imaging through FDGPET has yielded some clinically important information for patients with cancer of the pancreas. Surprisingly, however, no papers have been published (to our knowledge) using in vivo or in vitro MRS to study the human pancreas that has been afflicted with malignancy.

259

There are some investigations of human pancreatic cancer cells implanted in experimental animals. Kaplan et al. [60] found that pancreatic tumors evaluated using proton MRS in rats showed elevated taurine and lactate levels compared to normal pancreas, with creatine and glutamate low in pancreatic neoplasms.

15.7 In Vitro MRS applied to some other cancers

There has been some application of in vitro MRS to other cancers, such as malignant melanomas and a variety of thyroid neoplasms. 15.7.1 Malignant melanoma Melanoma metastases to lymph nodes have been analyzed using proton MRS by Lean et al [61]. The ratio of 1.8 to 2.5 ppm (containing lipid, lactate and other metabolites) to choline (3.2 ppm) was significantly higher in excised benign lymph nodes compared to those containing melanoma. Thompson et al. [62] suggest that techniques such as in vivo proton MRS hold great promise for assessment of sentinel nodes in patients with melanoma (p. 147S). 15.7.2 Cancer of the thyroid Russell et al. [63] performed 1D and 2D in vitro proton MRS analysis of thyroid tissue from 53 patients who underwent partial or total thyroidectomy for solitary thyroid nodules. The major differences between normal thyroid tissue and papillary carcinomas were the absence of lipid (CH3 resonates at 0.86 ppm) and the presence of amino acid metabolites (methyl resonances from amino acids resonate at 0.9 ppm) in cancerous tissue. The ratio of the resonance at 1.7 ppm to that at 0.9 ppm was > 1.1 in all the normal thyroid specimens, whereas in all of the thyroid cancers (papillary, medullary and anaplastic) the ratio was < 1.1 and there was no overlap between normal and malignant tissue. With respect to the follicular neoplasms, those which were histologically or clinically clearly malignant all had a ratio of 1.7 ppm to 0.9 ppm < 1.1. This group of authors in a subsequent paper [43] illustrated the importance of MRSI applied in vitro to the thyroid, demonstrating the chemical heterogeneity of follicular thyroid neoplasms, which are morphologically homogeneous.

References
[1] J. Czernin, M.E. Phelps, Positron emission tomography scanning: Current and future applications, Annu. Rev. Med. 53, 89-112 (2002).

260

[2] M. Phelps, E. Hoffmann, N. Mullani, M. Ter-Pogossian, Application of annihilation coincidence detection to transaxial reconstruction tomography, J. Nucl. Med. 16, 210-224 (1975). [3] M. G. Pomper, Functional and metabolic imaging, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 679-689. [4] K. Kinkel, Y. Lu, M. Both et al., Detection of hepatic metastases from cancers of the gastrointestinal tract by using non-invasive imaging methods (US, CT, MR imaging, PET): A meta-analysis, Radiology 224, 748-756, (2002). [5] A. Annovazzi, M. Peeters, A. Maenhout, A. Signore, R. Dierckx, C. Van de Wiele, 18flurodeoxyglucose positron emission tomography in nonendocrine neoplastic disorders of the gastrointestinal tract, Gastroenterology 125, 1235-1245 (2003). [6] J.D. Potter, D. Hunter, Colorectal cancer, in: H-O. Adami, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, Oxford, 2002, p. 188-211. [7] R.J. Mayer, Gastrointestinal tract cancer, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 578-588. [8] J.M. Skibber, B.D. Minsky, P.M. Hoff, Cancer of the colon, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 1216-1271. [9] B. Saar, T. Rsch, E.J. Rummeny, Colorectal cancer screening: A challenge for magnetic resonance colonography, Top. Magn. Reson. Imaging 13, 427-434 (2002). [10] G.N. Tzimas, D.J. Koumanis, S. Meterissian, Positron emission tomography and colorectal carcinoma: an update, J. Am. Coll. Surg. 198, 645-652 (2004). [11] M.E.J. Pijl, A.S. Chaoui, R.L. Wahl, J.A. van Oostayen, Radiology of colorectal cancer, Eur. J. Cancer 38, 887-898 (2002). [12] A. Oto, Virtual endoscopy, Eur. J. Radiol. 42, 231-239 (2002). [13] P.B. Cotton, V.L. Durkalski, B.C. Pineau, et al., Computed tomographic colonography (virtual colonoscopy) A multicenter comparison with standard colonoscopy for detection of colorectal neoplasia, JAMA 291, 1713-11719 (2004). [14] M.S. Karpeh, D.P. Kelsen, J.E. Tepper, Cancer of the stomach, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p.1092-1126. [15] D.J. Blezek, R.A. Robb, Evaluating virtual endoscopy for clinical use, J. Digital Imaging 103 (Suppl. 1), 51-55 (1997). [16] S. Mazzeo, D. Caramella, A. Gennai, et al., Multidetector CT and virtual endoscopy in the evaluation of the esophagus, Abdom. Imaging 29, 2-8 (2004). [17] M.K. Kalra, M.M. Maher, P.R. Mueller, S. Saini, State-of-the-art imaging of pancreatic neoplasms, Br. J. Radiol. 76, 857-865 (2003). [18] N.R. Maisey, A. Webb, G.D. Flux, et al., FDG-PET in the prediction of survival of patients with cancer of the pancreas: a pilot study, Br. J. Cancer 83, 287-293 (2000). [19] A.M. Xu, H.Y. Cheng, W.B. Jiang, D. Chen, Y.C. Jia, M.C. Wu, Multi-slice threedimensional spiral CT cholangiography: a new technique for diagnosis of biliary diseases, Hepatobil. Pancreat. Dis. Int. 1, 595-603 (2002).

261

[20] A.E. Li, D.A. Bluemke, Magnetic Resonance Imaging in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 669-679. [21] M.R. Arguedas, Screening for hepatocellular carcinoma: why, when, how? Gastroenterol. Rep. 5, 57-62 (2003). Curr.

[22] S.C.H. Yu, D.T.K. Yeung, N.M.C. So, Imaging features of hepatocellular carcinoma, Clin. Radiol. 59, 145-156 (2004). [23] R. Soper, U. Himmelreich, D. Painter, et al., Pathology of hepatocellular carcinoma and its precursors using proton magnetic resonance spectroscopy and a statistical classification strategy, Pathology 34, 417-422 (2002). [24] T. Nishijima, M. Nishina, K. Fujiwara, Measurement of lactate levels in serum and bile using proton nuclear magnetic resonance in patients with hepatobiliary diseases: its utility in detection of malignancies, Jpn. J. Clin. Oncol. 27, 13-17 (1997). [25] R.M. Dixon, NMR studies of phospholipid metabolism in hepatic lymphoma, NMR Biomed. 11, 370-379 (1998). [26] K. Takahama, Y. Amano, H. Hayashi, M. Ishihara, T. Kumazaki, Detection and characterization of focal liver lesions using superparamagnetic iron oxide-enhanced magnetic resonance imaging: comparison between ferumoxides-enhanced T1-weighted imaging and delayed-phase gadolinium-enhanced T1-weighted imaging, Abdom. Imaging 28, 525-530 (2003). [27] G. Brinkmann, U.H. Melchert, L. Emde, et al., In vivo P-31-MR-spectroscopy of focal hepatic lesions. Effectiveness of tumor detection in clinical practice and experimental studies of surface coil characteristics and localization technique, Investig. Radiol. 30, 56-63 (1995). [28] I.R. Francis, T.L. Chenevert, B. Gubin, et al., Malignant hepatic tumors: P-31 MR spectroscopy with one-dimensional chemical shift imaging, Radiology 180, 341-344 (1991). [29] I.J. Cox, J.D. Bell, C.J. Peden, et al., In vivo and in vitro 31P magnetic resonances spectroscopy of focal hepatic malignancies, NMR Biomed. 5, 114-120 (1992). [30] M. Ljungberg, G. Westberg, B. Vikhoff-Baaz, et al., P-31 MRS of liver metastases of neuroendocrine tumors treated with hepatic artery embolisation, MAGMA 15 (Suppl. 1), 130-131 (2002). [31] D.J. Meyerhoff, G.S. Karczmar, F. Valone, A. Venook, G. B. Matson, M.W. Weiner, Hepatic cancers and their response to chemoembolization therapy. Quantitative image-guided 31P magnetic resonance spectroscopy, Investig. Radiol. 27, 456-464 (1992). [32] A. Schilling, B. Gewiese, G. Berger, et al., Liver tumors: follow-up with P-31 MR spectroscopy after local chemotherapy and chemoembolization, Radiology 182, 887-890 (1992). [33] K. Taniguchi, T. Kaminaga, M. Sakon, et al. The change of hepatic energy status after transcatheter arterial embolization (TAE) for hepatocellular carcinomaa study using 31P.MRS, Jpn. J. Cancer Chemother. 24, 1632-1634 (1997). [34] J.D. Bell, I.J. Cox, J. Sargentoni, et al., A 31P and 1H-NMR investigation in vitro of normal and abnormal human liver, Biochem. Biophys. Acta 1225, 71-77 (1993). [35] T.H. Saunders, H.K. Mendes Ribeiro, F.V Gleeson, New techniques for imaging colorectal cancer: the use of MRI, PET and radioimmunoscintigraphy for primary staging and follow-up, Br. Med. Bull. 64, 81-99 (2002). [36] W. Luboldt, P. Bauerfeind, S. Wildermuth, B. Marincek, M. Fried, J.F. Debatin, Colonic masses: Detection with MR colonography, Radiology 216, 383-388 (2000).

262

[37] G. Pappalardo, E. Polettini, F.M. Frattaroli, et al., Magnetic resonance colonography versus conventional colonoscopy for the detection of endoluminal lesions, Gastroenterology 119, 300304(2000). [38] W. Luboldt, M.M. Morrin, MR colonography: status and perspective, Abdom. Imaging 27, 400-409 (2002). [39] D.F. Ransohoff, Virtual colonoscopywhat it can do vs. what it will do, JAMA 291, 17721774 (2004). [40] J. A. Koutcher, M. Goldsmith, R. Damadian, NMR in cancer. X. A malignancy index to discriminate normal and cancerous tissue, Cancer 41, 174-182 (1978). [41] P. Kowalski, P. Skupin, J. Sowier, K.J. Olszewski, NMR relaxation time studies of large bowel neoplasms, Physiol. Chem. Phys. Med. NMR 29, 51-54 (1997). [42] C.E. Mountford, G.L. May, P.G. Williams et al., Classification of human tumors by highresolution magnetic resonance spectroscopy, Lancet 1 (8482), 651-653 (1986). [43] C.E. Mountford, W.A. MacKinnon, P. Russell, A. Rutter, E.J. Delikatny, Human cancers detected by proton MRS and chemical shift imaging ex vivo, Anticancer Res. 16, 1521-1532 (1996). [44] A. Moreno, M. Rey, J.M. Montane, J. Alonso, C. Arus, 1H NMR spectroscopy of colon tumors and normal mucosal biopsies: elevated taurine levels and reduced polyethyleneglycol absorption in tumors may have diagnostic significance, NMR Biomed. 6, 111-118 (1993). [45] A. Moreno, C. Ars, Quantitative and qualitative characterization of 1H NMR spectra of colon tumors, normal mucosa and their perchloric acid extracts: decreased levels of myoinositol in tumors can be detected in intact biopsies, NMR Biomed. 9, 33-45 (1996). [46] J.N. Kasimos, T.E. Merchant, L.W. Gierke, T. Glonek, 31P magnetic resonance spectroscopy of human colon cancer, Cancer Res. 50, 527-532 (1990). [47] T.E. Merchant, J.N. Kasimos, P.W. de Graaf, B.D. Minsky, L.W. Gierke, T. Glonek, Phospholipid profiles of human colon cancer using 31P magnetic resonance spectroscopy, Int. J. Colorectal Dis. 6, 121-126 (1991). [48] T.E. Merchant, P.M. Diamantis, G. Lauwers, et al., Characterization of malignant colon tumors with 31P nuclear magnetic resonance phospholipid and phosphatic metabolite profiles, Cancer 76, 1715-1723 (1995). [49] E. Krupnik, K.M. Briere, R.P. Bird, C. Littman, I.C. Smith, 1H magnetic resonance spectroscopy evidence that aberrant crypt foci are preneoplastic lesions in the colon, Anticancer Res. 19, 1699-1704 (1999). [50] K. Nakagami, T. Uchida, S. Ohwada, et al., Increased choline kinase activity and elevated phosphocholine levels in human colon cancer, Japan. J. Cancer Res. 90, 419-424 (1999). [51] T. Motohara, R.C. Semelka, MRI in staging of gastric cancer, Abdom Imaging 27, 376-383 (2002). [52] F. Maccioni, Current status of gastrointestinal MRI, Abdom. Imaging 27, 358-360 (2002). [53] C.W. Mun, J.Y. Cho, W.J. Shin, et al. Ex vivo proton MR spectroscopy ((1) H-MRS) for evaluation of human gastric carcinoma, Magn. Reson. Imaging 22, 861-870 (2004). [54] U. R. Dave, A.D. Williams, J.A. Wilson et al., Esophageal cancer staging with endoscopic MR imaging: pilot study, Radiology 230, 281-286 (2004). [55] A. Giovognoni, G. Valeri, C. Ferrara, MRI of esophageal cancer, Abdom. Imaging 27, 361366 (2002).

263

[56] S.T. Doran, G.L. Falk, R.L. Somorjai, et al., Pathology of Barretts esophagus by proton magnetic resonance spectroscopy and a statistical classification strategy, Am. J. Surg. 185, 232238 (2003). [57] T.E. Merchant, P.W. de Graaf, B.D. Minsky, H. Obertop, T. Glonek, Esophageal cancer phospholipid characterization by 31P NMR, NMR Biomed. 6, 187-193 (1993). [58] T.E. Merchant, B.D. Minsky, G.Y. Lauwers, P.M. Diamantis, T. Haida, T. Glonek, Esophageal cancer phospholipids correlated with histopathologic findings: a 31P NMR study, NMR Biomed. 12, 184-188 (1999). [59] Y. Tanizawa, T. Nakagohri, M. Konishi, et al., Virtual pancreatoscopy of pancreatic cancer, Hepato-Gastroenterol. 50, 559-562 (2003). [60] O. Kaplan, T. Kushnir, N. Askenazy, T. Knubovets, G. Navon, Role of nuclear magnetic resonance spectroscopy (MRS) in cancer diagnosis and treatment: 31P, 23Na and 1H MRS studies of three models of pancreatic cancer, Cancer Res. 57, 1452-1459 (1997). [61] C.L. Lean, R. Bourne, J.F. Thompson, et al., Rapid detection of metastatic melanoma in lymph nodes using proton magnetic resonance spectroscopy of fine needle aspiration biopsy specimens, Melanoma Res. 13, 259-261 (2003). [62] J.F. Thompson, J.R. Stretch, R.F. Uren, V.S. Ka, R.A. Scolyer, Sentinel node biopsy for melanoma: Where have we been and where are we going? Ann. Surg. Oncol. 11 (Suppl.), 147S151S (2004). [63] P. Russell, C.L. Lean, L. Delbridge, G.L. May, S. Dowd, C.E. Mountford, Proton magnetic resonance and human thyroid neoplasia I: Discrimination between benign and malignant neoplasms, Am. J. Med. 96, 383-388 (1994).

265

Chapter 16

Breast Cancer: Screening and Early Detection

_______________________________________________________________________________

16.1

Overview of epidemiological & clinical aspects

Incidence and prevalence/morbidity and mortality

Worldwide, breast cancer is the most commonly occurring cancer in women, and the leading cause of cancer-related deaths among women. In the year 1990, there were an estimated 795 600 newly diagnosed cases of breast cancer worldwide [1]. Breast cancer rates vary greatly worldwide, with the highest rates in North America and Europe; about one-fifth to one-tenth the occurrence is seen in Asia [2]. Within the U.S. the incidence rates are increasing the fastest among Hispanic women [3]. Women of Ashkenazi Jewish descent are known to be at elevated risk due to the high prevalence (approximately 2.5%) of founder mutations in the breast cancer susceptibility gene BRCA 1 [4]. Icelandic populations are reported to have a high prevalence of BRCA 2 mutations [1] (see next subsection). On the other hand, the relatively low risk in Asia cannot be explained by genetic factors, since Asian women who live in western countries have similar occurrence of breast cancer as those in their western counterparts [2]. Overall, there is an increasing breast cancer incidence with age, but the rapid rise in rates slows to some extent around age 50 [1]. The gender-related risk of breast cancer is about 150:1 for females; this means, of course, that while breast cancer is predominantly a disease of women, it can also affect men [2].

266

Etiology/risk factors Inherited susceptibility An estimated 5-10% of breast cancers can be linked directly to highly penetrant germline mutations. Having a mother or sister with breast cancer confers a relative risk of 1.5 to 3.0 compared to those without a first-degree positive family history. For women with both mother and a sister who have had breast cancer, in particular at a young age, the risk is much higher [1].

BRCA 1 and BRCA 2 These genes are presumed to act as tumor suppressors. An inherited mutated allele on BRCA 1 confers a 60-80% lifetime chance of developing breast cancer (as well as a 33% chance of developing ovarian cancer). Germline mutations in BRCA 2 are associated with an increased risk of breast cancer in men as well as in women [2]. Ataxia-telangiectasia Autosomal recessive, about 1.4% of the general population is heterozygous, for AT, with heightened vulnerability to DNA damage from ionizing radiation, chemotherapy and hydroxy radicals. The relative risk of breast cancer among heterozygotes is estimated at 3.9 (95% CI = 2.1- 7.2) [1]. Li-Fraumeni Syndrome Inherited mutations in the p53 tumor suppressor gene are associated with an increased risk of breast cancer, as well as sarcomas and other malignancies, as noted. Cowdens diseaseMultiple Hamartoma Syndrome Autosomal dominant deletion of the PTEN1 tumor-suppressor gene is associated with a 30-50% risk of breast cancer by age 50. This is often bilateral [1].

A number of low-penetrance polymorphisms, related mainly to the metabolism of hormones and carcinogens have been identified, which confer an increased risk of breast cancer [1]. Gynecologically related risk factors
Early menarche, late menopause, nulliparity Women who have a first full term pregnancy by age 18 have about a 30 to 40% decreased risk of breast cancer compared to nulliparous women. Menarche at age 16 is associated with 50 to 60% of the risk of those with menarche at age 12. Menopause at age 42 confers a 35% lower risk of breast cancer compared to menopause at age 52. These findings are all considered to
1

PTEN is the acronym for phosphatase and tensin homolog deleted on chromosome 10

267

be related to endogenous estrogen, since women without functioning ovaries who never receive estrogen replacement do not develop breast cancer (p. 572) [2]. Combined estrogen-progestin HRT Use even for relatively short periods is associated with an increased risk of breast cancer, as recently confirmed by randomized clinical trials such as the Womens Health Initiative [5]. Current or recent use of oral contraceptives While the data are not entirely consistent, there is substantial evidence that current or recent use of oral contraceptives confers a modestly increased risk of breast cancer [1]. Benign breast disease Benign proliferative disease of the breast is associated with a relative risk of about 1.5 to 2.0 for breast cancer. The relative risk for atypical hyperplasia is 4.0 to 5.0 [6].

Lifestyle related factors

Obesity The evidence is consistent for women in later adult life. This may be related to elevated circulating estrogens derived from adipose tissue [1, 6]. There is also evidence that among women prior to menopause, central, but not general obesity may increase the risk of breast cancer [7]. Low levels of physical activity Sedentary lifestyle is a likely risk factor, although the data are not entirely consistent [1]. A recent case-control study examining lifetime physical activity reports that for women prior to menopause the multivariate adjusted OR was 0.74 (95% CI = 0.52 1.05) for the highest versus lowest tertile of average lifetime activity, and 0.81 (95% CI = 0.64 1.02) for women after menopause [8]. Alcohol A dose response relationship has been reported between ethanol consumption and risk of breast cancer. From meta-analysis, it is reported that for 1 drink per day the relative risk is 1.1, rising to 1.2 for 2 drinks per day. The relative risk is 1.4 for 3 or more drinks per day [6]. The association between higher alcohol consumption is seen for in situ, localized as well as regional breast cancer [9].

Other Medical Conditions and Related Exposures

Exposure to radiation Exposure to radiation from multiple fluoroscopies or for treatment of Hodgkins lymphoma before the age 30 is associated with increased risk of breast cancer [2]. Follow-up of persons exposed to moderate to high levels of

268

ionizing radiation consistently demonstrates a significantly increased risk of breast cancer, and the magnitude of risk is inversely related to age of exposure [1].

Occupational and Environmental Factors Data concerning specific occupations and risk of breast cancer have not generally been consistent; studies have not usually been adjusted for confounders, and have most often been based on job titles rather than exposures. Insight into work-related factors that may be important in the etiology of breast cancer is provided by prospective, rigorously designed investigations within an occupation.
Night-shift work A recent 10-year prospective study of 78 562 female nurses [10] revealed that nurses who had worked at least 3 night shifts per month for over 30 years had a significantly increased risk of breast cancer compared to nurses who had not ever worked at least 3 night shifts per month. The relative risk was 1.36 (95% CI 1.04 1.78) after adjusting for age, age at menarche, parity, age at first birth, body mass index at age 18, family history of breast cancer (sister or mother), benign breast disease, oral contraception, alcohol, time, age at menopause, post-menopausal hormone use and height. There have been a number of suggested mechanisms for increased breast cancer risk associated with night shift work. The strongest biological evidence is related to the melatonin pathway, which is suppressed by artificial light. Melatonin has antiproliferative activity and modulates effects on estrogen receptors [10, 11]. The finding that blind women have lower risk of breast cancer is also broadly corroborative with this proposed mechanism [11].

Clinical Presentation and Differential Diagnosis Nowadays, before any signs or symptoms are noted, many breast cancers are detected on screening mammography (see next Section, 16.2). In the past, a palpable (usually non-tender) mass was the typical presentation of breast cancer. Other changes such as dimpling of the breast, flattened or inverted nipple, bloody or clear discharge from the breast, axillary lymphadenopathy, peau d-orange2 also could be manifested [12]. Hard, irregular, tethered or fixed non-tender lesions are more characteristic of breast cancer. However, lack of these physical signs is not sufficiently reliable to rule out malignancy [2]. The differential diagnosis of a solitary non-tender breast mass includes:
2

Cyst of the breast Fibroadenoma

Peau d orange is defined as deeply indented skin with holes that are the accentuated orifices of sweat glands; this is due to lymphatic obstruction producing edema of the breast (p. 256) [12].

269

Chronic breast abscess Fat necrosis of the breast Tuberculosis of the breast.

Breast carcinoma may present as an acute inflammation, especially in the lactating breast. The differential diagnosis would then include: Acute mastitis Acute abscess of the breast.

While the initial presentation is most often a single mass, more than 1 locus can be affected by breast cancer. The differential diagnosis of multiple breast masses includes: Chronic cystic mastitis Fibroadenosis Galactocele (during or after lactation).

Breast cancer during pregnancy As stated by Lippman [2]: development of a dominant mass during pregnancy or lactation should never be attributed to hormonal changes (p. 574). It is estimated that breast cancer develops in 1 of 3 000 to 4 000 pregnancies. When diagnosed, pregnant women frequently have more advanced disease because a breast mass was ignored (p. 574) [2]. Classification and Grading and Associated Prognosis Most breast cancers are adenocarcinomas. Expression of estrogen receptor is associated with improved prognosis. Over-expression of HER-1/neu, p53 mutations, high growth fraction and aneuploidy are poor prognostic features. Staging The following are staging criteria based on the TNM classification, and corresponding prognosis, as presented in Ref. [2].
Primary Tumor T0 = No evidence of primary tumor Tis = Carcinoma in situ T1 = Tumor 2cm

270

T2 = Tumor > 2 cm, but 5 cm T3 = Tumor > 5 cm T4 = Tumor extends into chest wall, satellite lesions, ulcerations. Regional Lymph Nodes N0 = No regional nodes N1 = Moveable ipsilateral lymph node involvement N2 = Matted or fixed ipsilateral lymph node involvement N3 = Ipsilateral internal mammary nodes. Distant Metastases M0 = No distant metastases M1 = Distant metastases (ipsilateral supraclavicular nodes are included here). Stage grouping Stage 0 = TIS N0 Stage I = T1 T0 T1 T2 T2 T3 T0 T1 T2 T3 T4 Any T Any T N0 N1 N1 N0 N1 N0 N2 N2 N2 N1,N2

M0 M0 M0 M0 M0 M0 M0 M0 M0 M0 M0

Stage IIA =

Stage IIb =

Stage IIIA =

Stage IIIb =

Any N M0 N3 M0 Any N M1.

Stage IV=

Percent 5-year survival by stage Stage 0 99% Stage I 92% Stage IIA 82% Stage IIB 65% Stage IIIA 47% Stage IIIB 44% Stage IV 14%.

Treatment
Treatment of breast cancer depends upon the stage of disease. Stage 0, (in situ ductal carcinoma) is treated with wide excision and RT, and

271

possibly adjuvant tamoxifen. This is also the approach to those in Stage I with tumors smaller than 1 cm. Adjuvant chemotherapy for 6 months is generally given in addition for higher stages. For estrogenpositive tumors larger than 1 cm, tamoxifen is added. It is estimated that breast cancer will recur in approximately half of the patients with localized disease. The reader is referred to Refs. [2, 6] and more recent clinical reviews for further information on therapeutic approaches to breast cancer.

16.2 Breast cancer screening with mammography

The currently recommended method for reducing breast cancer mortality is through a clinical breast examination and mammogram. The American Cancer Society recommends that these begin at age 40 for women at average risk [13]. Mortality reduction associated with screening mammography Regular screening with mammography has been consistently shown in randomized controlled trials to provide long-term reduction in breast cancer mortality [13]. In order to achieve a 30% reduction in breast cancer mortality, 80% of women aged 50 to 70 should comply with these guidelines. However, compliance with screening guidelines is often inadequate to achieve this goal. This is particularly the case among women of lower socioeconomic levels and among ethnic minorities [14]. Problems with mammography and barriers to compliance with screening guidelines
Radiation exposure with mammography Concern over radiation from mammography represents one important barrier to compliance with screening recommendations [14-16]. This concern is especially relevant for women below the age of 50, where the balance between the number of breast cancer deaths prevented by screening compared with the number induced by radiation seems less favourable (p. 81) [17]. The female breast is a radiosensitive organ. As mentioned earlier in this chapter, exposure to ionizing radiation is consistently associated with an increased risk of breast cancer. The magnitude of risk is inversely related to age of exposure [1]. The very low energy X-rays used in screening mammography may be more harmful, per unit dose, than high-energy X-rays. Consequently, Brenner and colleagues [18] consider the estimated radiation risk for younger women as sufficient to warrant commencing routine screening 5-10 years later than currently recommended.

272

Increased radiosensitivity with hereditary risk of breast cancer Mutations in the BRCA genes or heterozygous AT status, result in impaired DNA repair mechanisms and heightened sensitivity to radiation [6]. At the same time, it is among these women with hereditary risk, for whom screening at an earlier age and at more frequent intervals than for women at average risk has been suggested as a possible option [13]. According to Kuni [19], among the very radiosensitive subgroup: the women bearing a mutation of the gene BRCA 1 or BRCA 2 repeated X-ray use must be definitely avoided (p. 443). Limitations in sensitivity of mammography Ros et al. [20] estimate that the overall false negative rate for mammography is about 10 to 15%. They note that up to 10% of breast cancers are not identified by mammography even when palpable. A false negative mammogram contributes to a delay in breast cancer diagnosis and results in poorer prognosis [21]. Moreover, limitations in sensitivity contribute to the perception that mammography is ineffective for the early detection of breast cancer, and may, consequently, lower compliance with screening guidelines [22]. Mammographically dense breasts Although mammography is very sensitive for detecting breast cancer in fatty breast tissue, detection of malignant lesions (unless calcified) by mammography is very difficult in dense breasts [23]. Dense breasts are commonly seen in younger women. This is another reason why early mammography screening for women at high risk can be problematic. Mammographically dense breasts are also associated with use of combined HRT and are considered a marker for increased breast cancer risk among women after menopause [24]. Non-calcified lesions Since breast cancer is a heterogeneous disease, the sensitivity of mammography varies in relation to the histopathologic subtype. While calcifications are relatively easy to see, these are present in only a minority of histologically-verified breast cancers, with the rest being non-calcified stellate and circular masses that are often much harder to perceive [25]. Low specificity of mammography Khalkhali and Itti [26] cite estimates of the positive predictive value of mammography as low as (15-30) %. Abnormal screening mammography may be followed up with compression or magnified views, and ultrasound can help determine whether the lesion is cystic or solid. Biopsy is recommended for breast lesions estimated to have 3% risk of cancer [2]. Up to 80 to 90% of breast biopsies show benign findings, with the vast majority being due to fibrocystic disease (small fluid-filled cysts and fibrous hyperplasia). As noted in Section 16.1, benign breast disease is associated with an increased risk of breast cancer [2]. Biopsies of benign lesions are associated with considerable morbidity, including: Anxiety.

273

The fear engendered by a false positive mammogram may lead to decreased compliance with screening guidelines, which can result in later-stage diagnosis of breast cancer and increased mortality [22]. Difficulties in subsequent mammographies after biopsy Once a breast biopsy has been performed, subsequent mammographic evaluation in the region of the scarred tissue is rendered more difficult [21].

Innovations in mammography & other X-ray based diagnostic methods Currently, screen-film mammography (SFM) is the gold standard for breast cancer screening. A number of promising innovations in mammography and other X-ray based methods for diagnosing breast cancer are on the horizon. Full-field digital mammography (FFDM) was recently approved by the U.S. Food and Drug Administration for breast cancer screening, and has shown improved specificity, i.e. significantly lower recall rate and lower biopsy rate compared to SFM. However, FFDM has an insignificantly lower sensitivity than SFM [13]. It should also be recalled that all these mammography-based techniques would still entail exposure to ionizing radiation.

16.3 Molecular imaging with FDG-PET and scintimammography

FDG-PET in primary diagnosis of breast cancer Thus far, FDG-PET has been applied for diagnosing primary breast carcinoma when mammographic findings were ambiguous. In a recent systematic review of the literature, Samson et al. [27] report an overall false negative rate of 12% for FDG-PET in the differential diagnosis of benign versus malignant lesions among patients with an abnormal mammogram of a palpable breast mass. Sensitivity was poorer for carcinoma in situ and multifocal lesions. The positive predictive value of FDG accumulation is reportedly better than 90% for breast cancer [20]. Scintimammography (SMM) Single photon scintimammography using 99mTc sestamibi, which is taken up by mitochondria, has also been applied for diagnosis of primary breast cancer, with an overall sensitivity and specificity of 83% and 81%, respectively. However, sensitivity is much lower for small tumors. In initial studies, dedicated SMM detectors have shown improved resolution, but these detectors also visualize more benign lesions [26].

274

16.4 Magnetic resonance: Early results in 1 breast cancer diagnosis

16.4.1 MRI With a number of recent advances, contrast-enhanced MRI has emerged as a method with sensitivity approaching 100% for detection of invasive breast cancer [28, 29]. MRI is particularly valuable for detecting malignant lesions in mammographically dense breasts [30]. Studies are currently on going in the U.S., Canada, Germany, Netherlands, U.K., France and Italy to evaluate MRI as a possible screening method for women at high risk for breast cancer [13]. Since most invasive breast cancers are hypervascular, they typically show intense contrast enhancement. MRI for detecting breast cancer relies almost exclusively upon the presence of neovascularity. This vasculature is leaky and contains many arterio-venous shunts. Thus, breast cancerous tissue enhances rapidly and then washes out over the next several minutes, whereas normal breast parenchyma enhances more slowly. These washout time intensity curves are typical of most (but not all) cancers. With administration of contrast agents such as gadolinium, lesions can be well visualized, especially with fat suppressed, T1 weighted imaging [28]. However, false negatives have been reported for welldifferentiated, invasive ductal carcinomas and for invasive lobular carcinomas. Moreover, the sensitivity of CE-MRI has been reported to be as low as 40% for ductal carcinoma in situ [28]. For more details on MRI findings in breast cancer, see Refs. [28-30]. While contrast-enhanced MRI has high spatial resolution and is more sensitive than mammography, it has limited specificity, thus sharing with mammography the problem of a high false positive rate (approximately 50%) [31]. Morris [28] cites the following as causes of false positive on MRI examinations of the breast: Fibroadenoma Ductal atypia Fibrocystic changes Papilloma Sclerosing adenosis Ductal hyperplasia.

275

16.4.2

H-MRS

Lippman [2] includes MRS among the newer techniques that may improve primary diagnosis of breast cancer. 1H MRS diagnostics based upon the presence or absence of a composite choline signal have been shown to increase the specificity of MRI with respect to the diagnosis of breast cancer [31]. Katz-Brull et al. [31] recently reviewed the five published clinical studies using single voxel in vivo 1H MRS, in which malignant and benign breast lesions were compared. They reported a sensitivity and specificity of 83% (95% CI = 73% - 89%) and 85% (95% CI = 71% - 93%), respectively, for identifying breast cancer in the 153 tumors examined, 53 of which were benign. Even better diagnostic accuracy was achieved among women aged 40 or younger, among whom there were 11 patients with breast carcinomas and 9 with benign breast lesions. The potential of 1H MRS for widespread application in breast cancer diagnostics was emphasized by these authors, provided that the factors limiting its diagnostic accuracy are overcome. Lipid suppression is particularly important in applications of 1H MRS to the breast, due to its high lipid content. The current strategy has been to increase the echo time, which diminishes the overlap with the lipid peak, although this is achieved by a diminution in signal intensity [31]. As a consequence, a smaller number of compounds are visualized and heretofore, the focus has been upon the composite choline peak. Choline may also be observed in benign breast lesions, as well as in the normal breast during lactation. Choline is often undetected in small tumors that are then misclassified as benign. As yet, the application of MRSI for breast cancer diagnostics has not been reported. 16.4.3 In vitro MRS findings
In contrast to in vivo 1H MRS breast examinations based mainly upon a single composite spectral entity (the total choline peak), the high resolution of in vitro MRS applied to extracted specimens provides a much greater insight into the metabolic activity of malignant breast tissue.

Biochemical pathways underlying the high choline in breast cancer

In vitro MR analysis of excised malignant breast tumors reveals that the composite choline peak contains a number of water-soluble metabolites such as phosphocholine, glycerophosphocholine, betaine and analogous compounds containing the ethanolamine head group and taurine, as well as choline itself. Milk, on the other hand, is comprised predominantly of choline compounds such as phosphatidylcholine as well as phosphocholine and free choline [31]. Katz-Brull et al. [32] applied in vitro analysis using tracer kinetics and 13C and 31 P MRS to examine the biochemical pathways underlying the high levels of water-soluble choline metabolites seen in breast cancer. They identified two

276

non-intersecting pathways: phosphorylation and oxidation of choline, to be augmented with malignant transformation of mammary cells, with increased synthesis of phosphocholine and betaine. They also found suppression of choline-derived ether lipids. A comparison of the metabolic characteristics of breast cancer, fibroadenoma and normal breast tissue using in vitro 1H MRS Gribbestad et al. [33] published a study using in vitro 1H MRS to compare fourteen extracts of malignant breast tissue and one fibroadenoma to noninvolved breast from the same group of patients. We performed detailed pairwise and logistic regression analysis of their data to ascertain the sensitivity and specificity of individual metabolite concentrations for identifying breast cancer [22].
Lactate and alanine in breast cancer, fibroadenoma and non-involved breast tissue

In Figure 16.1, the case-by-case lactate levels for normal versus malignant breast tissue are shown. It is clearly seen that lactate concentrations were much higher in the malignant lesions than in the normal tissue of each patient.

Figure 16.1: Case-by-case analysis of estimated [lactate] in normal versus cancerous breast tissue (From data by Gribbestad et al. [33])

10

0 Case 1 Case 3 Case 5 Case 7 Case 9 Case 11 Case 13 Case 15 Case 2 Case 4 Case 6 Case 8 Case 10 Case 12 Case 14 Case 16

LAC_N LAC_LE

Lac_N denotes [lactate] in non-involved breast tissue, Lac_LE denotes [lactate] in the malignant breast lesion

Using logistic regression analysis, we calculated the sensitivity and specificity of the estimated metabolic concentrations. In Table 16.1 we present the

277

findings for lactate and alanine, comparing the malignant to the entirely normal extracts (denoted with (A)), and malignant versus benign breast tissue (denoted with (B)) and including 1 fibroadenoma). It is seen that lactate concentrations provided 100% diagnostic accuracy, for distinguishing benign versus malignant breast tissue, both when the fibroadenoma was excluded as well as when it was included.

Table 16.1 Diagnostic accuracy of estimated [lactate] and [alanine] for identifying breast cancer
(Data from Gribbestad et al. [33], calculations from Refs. [22])

Sensitivity (A)

Specificity (A)

Sensitivity (B)

Specificity (B)

Lactate [ 1.33 ppm ]

100%

100%

100%

100%

Alanine [ 1.47 ppm ]

92.9%

100%

92.9%

100%

(A) Denotes Malignant (N=14) versus non-infiltrated breast tissue (N=12), fibroadenoma excluded (B) Denotes Malignant (N=14)) versus benign breast tissue (N=13) (non-infiltrated or in 1 case fibroadenoma)

_______________________________________________________________

However, as shown in Figure 16.2, in the same patient [lactate] in the fibroadenoma was about 1.5 times greater than in the normal breast tissue from the same patient. The lactate concentration in the fibroadenoma was 2.6 Sd higher than the mean in normal breast tissue.

278

_____________________________________________________________

Figure 16.2: Estimated [lactate] in fibroadenoma and in normal breast tissue from the same patient (From data by Gribbestad et al. [33])

1,6 1,4 1,2 1 0,8 0,6 0,4 0,2 0 Fibroadenoma Normal Breast Tissue MicroM/g

________________________________________________________

Total choline and its constituents in breast cancer, fibroadenoma and non-involved breast tissue Next we examined the diagnostic accuracy of the estimated concentrations of total choline and its constituents for identifying breast cancer. As shown in Table 16.2, choline had 100% sensitivity only when the fibroadenoma was excluded. With inclusion of the fibroadenoma, there were two false negatives based upon choline concentrations. Each of the other constituents of total choline had less than perfect sensitivity, while phosphocholine had 100% specificity. Notably, total choline yielded marginally poorer results than most of the choline constituents, except for specificity when the fibroadenoma was included.

279

Table 16.2 Diagnostic accuracy of estimated concentrations of total choline and its constituents for identifying breast cancer
(Data from Gribbestad et al. [33], calculations from Refs. [22])

Sensitivity (A)

Specificity (A)

Sensitivity (B)

Specificity (B)

Choline [ 3.21 ppm ]

100%

100%

85.7%

100%

Phosphocholine [ 3.22 ppm ]

92.9%

100%

92.9%

100%

Glycerophosphocholine [ 3.23 ppm ]

85.7%

91.7%

85.7%

92.3%

Total Choline (C+PC+GPC)

92.9%

91.7%

92.9%

100%

(A) Denotes Malignant (N=14) versus non-infiltrated breast tissue (N=12), fibroadenoma excluded (B) Denotes Malignant (N=14)) versus benign breast tissue (N=13) (non-infiltrated or in 1 case fibroadenoma)

_______________________________________________________________

Moreover, as shown in Figure 16.3, in the same patient [total choline] in the fibroadenoma was also about 1.5 times greater than in the normal breast tissue from the same patient. The total choline concentration in the fibroadenoma was 1.81 Sd higher than the mean in normal breast tissue.

280

Figure 16.3: Estimated [total choline] in fibroadenoma and in normal breast tissue from the same patient (From data by Gribbestad et al. [33])

0,2 0,18 0,16 0,14 0,12 0,1 0,08 0,06 0,04 0,02 0 Fibroadenoma

MicroM/g

Normal Breast Tissue

______________________________________________________________

Phosphethanolamine, -glucose, taurine and myoinositol in breast cancer, fibroadenoma and noninvolved breast tissue Finally, we present the analysis of phosphethanolamine, -glucose, taurine and myoinositol concerning diagnostic accuracy in distinguishing breast cancer, fibroadenoma and non-involved breast tissue. As shown in Table 16.3, none of these metabolites provided perfect sensitivity, and only phosphoethanolamine and taurine yielded 100% specificity with and without inclusion of the fibroadenoma.

281

Table 16.3 Diagnostic accuracy of estimated concentrations of phosphoethanolamine, -glucose, taurine and myoinositol for identifying breast cancer
(Data from Gribbestad et al. [33], calculations from Refs. [22])

Sensitivity (A)

Specificity (A)

Sensitivity (B)

Specificity (B)

Phosphoethanolamine [ 3.22 ppm ] -Glucose [ 3.25 ppm ]

92.9%

100%

92.9%

100%

16.7%

83.3%

16.7%

92.3%

Taurine [ 3.27 ppm ]

92.9%

100%

92.9%

100%

Myoinositol [ 3.28 ppm ]

66.7%

83.3%

66.7%

84.6%

(A) Denotes Malignant (N=14, except for -Glucose N=6 and for myoinositol N=9) versus noninfiltrated breast tissue (N=12), fibroadenoma excluded (B) Denotes Malignant (N=14, except for -Glucose N=6 and for myoinositol N=9)) versus benign breast tissue (N=13) (non-infiltrated or in 1 case fibroadenoma)

Compared to most of the other metabolites, myoinositol did not have very high overall diagnostic accuracy. However, myoinositol did offer some diagnostic insight that the other metabolites failed to provide. Namely, the calculated concentration of myoinositol was nearly the same (0.465 and 0.448) for the fibroadenoma and for the non-infiltrated tissue, respectively, from the same patient and showed the lowest difference from the mean for normal breast tissue (+ 0.52 Sd) (see Figure 16.4).

282

Figure 16.4: Estimated [myoinositol] in fibroadenoma and in normal breast tissue from the same patient, and mean [myoinositol] in malignant breast tissue and in normal breast (From data by Gribbestad et
al. [33])

1,4 1,2 Fibroadenoma 1 0,8 0,6 0,4 0,2 0

_______________________________________________________________

Normal Tissue Mean Malignant Breast Mean Normal Tissue

Overall assessment On the basis of these data from a fairly small sample (with substantial missing data for a few metabolite concentrations in malignant tissues), definitive conclusions cannot be drawn about which metabolites are optimal for detecting the presence of breast cancer and distinguishing this from normal breast tissue or from benign lesions. Nevertheless, several metabolites (lactate, in particular) showed promise with respect to diagnostic accuracy. On the other hand, total choline, upon which most in vivo 1H MRS diagnoses are based, had marginally lower sensitivity and specificity than several other metabolites. Furthermore, metabolites such as myoinositol also provided potentially important insights, even though the calculated sensitivity and specificity were less than optimal. Viewed together, these analyses corroborate this group of authors [33] that a very rich window of information is provided by in vitro 1H MRS analysis of metabolite concentrations in malignant versus non-cancerous breast tissue. This should justify exploration of how in vivo 1H MRS might tap into this rich source of information for improved primary breast cancer diagnostics.

283

References
[1] S. Hankinson, D. Hunter, Breast cancer, In: Adami H-O, Hunter D, Trichopoulos D. Textbook of Cancer Epidemiology, New York, NY: Oxford University Press, 2002, p. 301-339. [2] M.E. Lippman, Breast cancer, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001,p. 571-578. [3] National Cancer Institute, Screening mammograms, retrieved December 21, 2001 from www.cancernet.nci.nih.gov [4] E. Dagan, R. Gershoni-Baruch, Hereditary breast/ovarian cancerpitfalls in genetic counselling, Clin. Genet. 60, 310-313 (2001). [5] R.T. Chlebowski, S.L. Hendrix, R.D. Langer, et al., Influence of estrogen plus progestin on breast cancer and mammography in healthy post-menopausal women: the Womens Health Initiative Randomized Trial, JAMA 289, 3243-3253 (2003). [6] E.P. Winer, M. Morrow, C.K. Osborne, J.R. Harris, Malignant tumors of the breast, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 1651-1717. [7] M. Harvie, L. Hooper, A.H. Howell, Central obesity and breast cancer risk: a systematic review, Obes. Rev. 4, 157-173 (2003). [8] E.M John, P.L. Horn-Ross, J. Koo, Lifetime physical activity and breast cancer risk in a multiethnic population: the San Francisco Bay area breast cancer study, Cancer Epidemiol. Biomarkers Prev. 12, 1143-1152 (2003). [9] H.S. Feigelson, C.R. Jonas, A.S. Robertson et al., Alcohol, folate, methionine and risk of incident breast cancer in the American Cancer Society Cancer Prevention Study II Nutrition Cohort, Cancer Epidemiol. Biomarkers Prev. 12, 161-164 (2003). [10] E.S. Schernhammer, F. Laden, F.E. Speizer, et al. Rotating night shifts and risk of breast cancer in the women participating in the nurses health study, J. Natl. Cancer Inst. 93, 1563-1568 (2001). [11] G. Glickman, R. Levin, G.C. Brainard, Ocular input for human melatonin regulation: relevance to breast cancer, Neuro. Endocrinol. Lett. 23 (Suppl. 2), 17-22 (2002). [12] E.L. DeGowin, R.L. DeGowin, Bedside Diagnostic Examination, 3rd Edition, Macmillan Publishing Co., New York, 1976. [13] R. A. Smith, D. Saslow, K.A. Sawyer, et al. American Cancer Society guidelines for breast cancer screening: update 2003. CA. Cancer J. Clin. 53, 141-169 (2003). [14] L.Tern, Compliance with Mammography Screening Guidelines among Latinas: An Exploratory Study, [Doctoral Dissertation] University of Southern California School of Medicine, Department of Preventive Medicine, 2004. [15] L. Remennick, I have no time for potential troubles: Russian immigrant women and breast cancer screening in Israel, J. Immigr. Health 5, 153-163 (2003). [16] R. Bastani, A. Marcus, A. Maxwell, et al. Evaluation of an intervention to increase mammography screening in Los Angeles, Prev. Med. 23, 83-90 (1994). [17] P.M. Beemsterboer, P.G. Warmerdam, R. Boer, H.J. de Koning, Radiation risk of mammography related to benefit in screening programmes: A favorable balance? J. Med. Screen. 5, 81-87 (1998).

284

[18] D.J. Brenner, S.G. Sawant, M.P. Hande, et al., Routine screening mammography: how important is the radiation-risk side of the benefit-risk equation? Int. J. Radiat. Biol. 78, 1065-1067 (2002). [19] H. Kuni, I. Schmitz-Feuerhake, H. Dieckmann, Mammography screeningneglected aspects of radiation risks, Gesundheitswesen 65, 443-446 (2003). [20] C. Ros, J. Dose, N. Avril, Positron emission tomography for the diagnosis of breast cancer, Nucl. Med. Commun. 23, 613-618 (2002). [21] E. Prats, Symposium of breast imaging, Nucl. Med. Commun. 23, 607-608. (2002). [22] K. Belkic, Current dilemmas and future perspectives for breast cancer screening with a focus upon optimization of magnetic resonance spectroscopic imaging by advances in signal processing, Isr. Med. Assoc. J. 6, 610-618 (2004). K. Belkic, Magnetic resonance spectroscopic imaging in breast cancer detection: possibilities beyond the conventional theoretical framework for data analysis, Nucl. Instr. Meth. Phys. Res. A. 525, 313-321 (2004). [23] D.B. Kopans, Breast Imaging, 2nd Edition, Philadelphia, Lippincott-Raven Publishers, 1998. [24] G.A. Greendale, B.A. Reboussin, S. Slone, et al., Postmenopausal hormone therapy and change in mammographic density, J. Natl. Cancer Inst. 95, 30-37 (2003). [25] L. Tabr, Challenges in mammography from a clinical perspective. [Abstract] In: Imaging 2003: International Conference on Imaging Techniques in Subatomic Physics, Astrophysics, Medicine, Biology and Industry, 2003 June, Stockholm, Sweden. [26] I. Khalkhali, E. Itti, Functional breast imaging using the single photon technique. Nucl. Med. Commun. 23, 609-611 (2002). [27] D.J. Samson, C.R. Flamm, E.D. Pisano, N. Aronson, Should FDG PET be used to decide whether a patient with an abnormal mammogram or breast finding at physical examination should undergo biopsy? Acad. Radiol. 9, 773-783 (2002). [28] E.A. Morris, Breast cancer imaging with MRI, Radiol. Clin. N. Am. 40, 443-446 (2002). [29] S.G. Orel, M.D. Schnall, MR imaging of the breast for the detection, diagnosis, and staging of breast cancer, Radiology 220, 13-30 (2001). [30] M.K. Wiberg, Magnetic resonance imaging in breast diagnosis, [Doctoral Dissertation] Karolinska Institute, Center for Surgical Sciences, Division of Radiology, 2002. [31] R. Katz-Brull, P.T. Lavin, R.E. Lenkinski, Clinical utility of proton magnetic resonance spectroscopy in characterizing breast lesions, J. Natl. Cancer Inst. 94, 1197-1203 (2002). [32] R. Katz-Brull, D. Seger, D. Rivenson-Segal D, et al., Metabolic markers of breast cancer, Cancer Res. 62, 1966-1970 (2002). [33] I. S. Gribbestad, B. Sitter, S. Lundgren, et al., Metabolite composition in breast tumors examined by proton nuclear magnetic resonance spectroscopy, Anticancer Res. 19, 1737-1746 (1999).

285

Part C Future Perspectives for MRS and MRSI in Cancer Diagnostics

Chapter 17

Limitations of MRS / MRSI in Oncology: Relation to Reliance on the Conventional Framework for Data Analysis
_______________________________________________________________________________

As summarized thus far in this book, MRS and MRSI have shown great promise in relation to various aspects of tumor diagnostics, and important strides have been made in the most recent period with these molecular imaging modalities. However, it is also clear that further improvements are still needed. In this section we analyse many of the problems and dilemmas encountered using MRS and MRSI in oncology and their relation to reliance upon the conventional Fourier-based framework for data analysis. The generally accepted notion is that data analysis in magnetic resonance imaging and spectroscopy with biomedical signals should rely upon the most conventional theoretical framework known as the fast Fourier transform. As recently stated by Hennig [1] The mathematical procedure for converting the spatially encoded time variant signal into the frequency domain is the Fourier Transformation. Fourier-based reconstruction techniques therefore constitute the most natural way of generating an MR-image (p.36). In this chapter we discuss why the above-cited statement needs to be reevaluated, in light of the limitations of the FFT. For a brief

286

mathematical exposition of the FFT, see Chapter 4 and references cited therein.

17.1 Low Resolution of FFT

It has been noted [2] that one of the critical problems for MR imaging and spectroscopy is the need for long imaging times with insufficient resolution. This limitation is due to performing data analysis based exclusively upon the FFT, which is a low-resolution spectral/image estimator. Recall from Chapter 4, that using the FFT, a spectrum is % given as a single polynomial, F( k ) Fk with pre-assigned angular

% % frequencies, k , whose minimal separation min = 2 k/T is determined by the given epoch (or the total acquisition time) T. The % FFT spectrum is defined only on the Fourier grid points k , = 2 k/T where (k = 0,1,2,3 N-1) and N is the signal length. The main strategy applied in attempts to improve resolution has % been to increase T, and thereby to decrease min . As noted in Chapter 4, within the FFT a strategy known as zero filling is used as an attempt at quasi-interpolation. It has been pointed out that this does not improve resolution, but merely renders the shape spectrum more visually presentable [3]. In clinical practice, at larger T, the signal becomes heavily corrupted with background noise1. The reason is that envelopes of time signals, such as those observed in MRS, decay exponentially, so that the larger signal intensities are found early in the recording [3]. It is therefore advantageous to encode the time signal as rapidly as possible, i.e. to avoid long acquisition times at which mainly noise will be recorded. In other words, there are two mutually exclusive requirements and as a result, within the FFT, attempts to improve resolution lead to a worsening of the SNR. This conundrum is graphically illustrated in Figure 17.1.
Particular problem for breast cancer diagnostics For breast cancer diagnostics using MRS this is particularly troublesome, due to the need for lipid suppression. One of the current strategies has been to increase the TE, which diminishes the overlap with the lipid signal, but this is achieved by a diminution in signal intensity. Poor SNR was cited as one of the major reasons for false
1 An attempt to suppress this noise at the tail end of the FID is often made by performing apodization. This entails multiplying the acquired FID by a smoothly varying function prior to Fourier transformation. However, apodization leads to line broadening [4].

287

negative findings using 1H MRS to detect malignant breast lesions [5]. Moreover, some of the potentially informative metabolites for identifying breast cancer have short T2 relaxation times, and will have decayed at longer TE [6-7].

Figure 17.1 The relation between increase in acquisition time and signal-to-noise ratio (SNR) for a free induction decay (FID) or digitized time signal {cn} with 512 time points (n = 0 - 511), as seen in a time signal from MRS. From Ref. [8] with permission.

The FID is represented by the dark, rapidly oscillating thin full line. Noise, for purposes of this illustration only, is represented by a sine wave (full black line, of constant amplitude and periodic repetition). It can be seen that as time (abscissa) increases, the SNR dramatically worsens. The abscissa is integer n, which measures time in milliseconds. The ordinate is the signal intensity in arbitrary units.

17.2

Poor signal-to-noise ratio of FFT

As described above, when using the FFT attempts at improving resolution by increasing acquisition time inevitably lead to decreased SNR. There is another major reason for poor SNR using this conventional approach to signal analysis. Namely, the FFT is a linear transform and, as such, imports noise as intact from the measured time

288

domain data to the frequency domain. A spectral transform Fk is linear if the coefficients {an,k} of the transformation from the time to the frequency domain are independent of the signal points, as is precisely the case for the FFT.
N 1 n =0

Fk =

1 N

n,k n

(17.1)

(an,k = constants = e 2i nk N ) Limitations for applying MRS and MRSI due to poor SNR
Limitations for deep-seated organs

Due to poor SNR, methods such as MRS and MRSI have thus far been fairly limited in applications to deep-seated organs and/or those that engender motion artefact. Thus, as seen in Part B of this book, there have been relatively few studies using MRS or MRSI in the detection of malignant processes in the gastro-intestinal tract, kidney, liver, etc. Investigations of these organs with MRSI require the use of respiratory triggering or navigator scans [9] (see Chapters 2, 14 and 15). Clearly, a major hindrance to early detection of ovarian cancer by MRSI would also be related to poor SNR in this small organ. In choosing a type of cancer for their multi-centered study using MRS, Griffiths et al. [10] ruled out tumors of the bowels, lungs and kidney all of which were too difficult for routine 31P MRS {due to} motion and susceptibility artefacts and being deep within the body {such that} signal/noise ratio would be too low (p. 2087). SNR problems in breast cancer diagnostics

As noted, false negative findings with respect to MRS for diagnosing breast cancer have been mainly attributed to poor SNR. Technical problems have also been cited, related most often to motion artefact, which, in fact, is also within the realm of an SNR problem. Most importantly with respect to the possibilities of using proton MRS for early detection, it has been noted that small breast cancers were more likely to have an undetectable choline peak, compared to larger malignant lesions [5]. Moreover, the use of 31P MRS, which obviates the need for lipid (as well as water) suppression, is limited with respect to breast lesions, by the requirements that the tumor be approximately 10 times larger than with proton MRS to obtain the same SNR [5]. SNR issues in brain tumor diagnostics

A number of the problems with current applications of MRS and MRSI in brain tumor diagnostics are related to resolution and SNR issues. Particularly troublesome in this regard is the limited possibility of MRS and MRSI to

289

detect very small brain tumors [11], at the very time at which therapeutic interventions would have the best chance for success. As pointed out by Huang et al. [12] in vivo MRS has low SNR, which severely limits capabilities to determine critical characteristics such as the grade of brain tumors. Attempts to improve the SNR have most often entailed either increasing the acquisition time, or increasing the volume of tissue from which data is acquired. The latter approach frequently results in a heterogeneous voxel with a mixture of tissue types. SNR problems with MRSI

There are special problems in achieving adequate SNR using MRSI within scan times of reasonable length [4]. Because of the vital importance of achieving volumetric coverage of tumors, which are often heterogeneous, the SNR issues specifically related to MRSI are of particular concern for oncology.

17.3 The FFT supplies only a shape spectrum

The FFT is a non-parametric estimator, which provides only the shape of spectral structures, but not quantification. The peak parameters are extracted by fitting the obtained structures to a sum of Lorentzians or Gaussians, or both, or so-called Voigt profiles [13]. Thus, much information that is contained in the signal is not obtained, such as the actual position, width, height and phase of each metabolite. As stated by Danielsen and Ross [14] quantification of metabolite concentrations permits a more profound clinical interpretation at the level of cellular pathology (p. 11). However, as a consequence of difficulties in quantification, applications of MRS/MRSI in oncology have often relied upon either a dichotomous variable (e.g. presence or absence of a composite choline peak) or upon ratios. Problems related to semi-quantitative approaches With respect to applications of in vivo 1H MRS for breast cancer detection, these have mainly been based upon a dichotomous variable, the presence or absence of a composite choline peak. This compromises diagnostic accuracy, since: (1) choline may also be observed in benign breast lesions and in the normal breast during lactation, and (2) choline is often undetected in small tumors that are then misclassified as benign [5]. A semi-quantitative assessment of metabolites provided some, but often limited diagnostic insight within in vivo studies, e.g. of ovarian tumors (Chapter 10, see Table 10.2) and brain tumors (Chapter

290

8, see Tables 8.3 and 8.4), and head and neck tumors (Chapter 11), inter alia, where misclassifications were also reported. Problems related to reliance upon metabolite ratios Because of the problems in absolute quantification, MRS-based tumor diagnostics, especially brain tumors, has often relied upon metabolite ratios (see Table 8.2).
Metabolite ratios are often dependent on TE Metabolite ratios are affected by measurement parameters. Since different metabolites have different relaxation times, metabolite ratios are also dependent upon TE. Changes in TE greatly impact upon the usefulness of metabolite ratios for distinguishing various grades of brain tumors [15]. Denominators in ratios can be affected by nonmalignant processes The use of ratios is problematic for accurate assignment of spectral changes to specific disease processes [16], since the denominator often varies for reasons unrelated to the oncologic process of interest. For example, the ratio of choline to NAA, upon which detection of brain tumor is heavily based in 1H MRS, may also be due to loss of NAA, which is seen in a fairly wide variety of neurological disorders including epilepsy, multiple sclerosis, as well as in cerebrovascular accidents.
Contamination from adjacent tissues

Metabolite ratios can be affected by contamination from adjacent tissues. For example, high NAA can occur with contamination from adjacent tissues within brain tumors [17].
Normal regional differences

Metabolite ratios can be affected by regional differences. For example, there is a wide variation in metabolic ratios in normal brain, especially comparing white and grey matter [11, 14]. In the prostate, normal citrate concentration varies substantially between the central gland and peripheral zone (see Chapter 9, Table 9.3).
Treatment related changes in metabolite ratios

Treatment-related changes can affect metabolite ratios. Alteration in the ratios of choline to NAA and to creatine may also occur as a result of sub-clinical CNS neurotoxicity related to chemotherapy, possibly due to neuronal loss and/or inhibition of cell metabolism, as well as to radiation effects [18, 19]. Hormone deprivation in prostate cancer treatment leads to a near total loss of observable metabolites, including citrate (see Chapter 9), such that the choline-citrate ratio cannot be used in these circumstances [20].

Thus, reliance on a ratio, where a change in the denominator could be due to a non-malignant process, especially one directly related to the

291

therapy, would severely hamper the ability to use MRS and MRSI to monitor response of tumor to therapy.

17.4 FFT requires for fitting, which is non-unique, the number of metabolites are guessed in advance
One of the key obstacles to a greater use of MRS in clinical practice has been the lack of a uniform approach to data analysis. Ars [21] has emphasized the need for data compatibility in order to make multicenter comparisons, indispensable for wider routine application of MRS in oncology. Danielsen and Ross [14] consider differences in data processing methods, rather than real differences, are probably the most important contributor to deviations between different clinical results
(p. 12).

At loggerheads with the need for data compatibility is the requirement within the FFT for fitting, which can lead both to spurious peaks (over-fitting) and true metabolites being undetected (underfitting). This is not only unacceptable to diagnosticians, but renders inter-study comparisons tenuous, at best, unless the same in vitro basis set is used in e.g. the LCModel [13], to predetermine the number of metabolites. Such fitting is often based upon priorknowledge/measurement of in vitro data, prior to analysis of the actual in vivo spectrum. Although claims have been made that fitting procedures can be automatic and objective [13], their major pitfalls and inherent subjectivity have been highlighted [3, 22]. As recently pointed out, if one does not include an in vitro metabolite in the basis set, then one is going to have a very bad fit precisely at the frequency location where the missing metabolite is expected to occur in the studied in vivo spectrum. This leads to subjectivity [23]. These problems are particularly pronounced with respect to overlapping metabolites. Various fitting procedures such as AMARES [24], MRUI [25], as well as the LCModel can be applied, none of which are generated with certainty, especially for the guessed number of metabolites and peak heights. Problems with Assignment/The Non-uniqueness of Fitting in Tumor Diagnostics Numerous problems with assignment and dilemmas related to the nonuniqueness of fitting have arisen with respect to tumor diagnostics, especially for brain tumors. These become particularly troublesome for closely situated and overlapping resonances. For example, with respect to brain metabolites, most authors have attributed peaks between 3.8 and 4.0 ppm to glutamine in the alpha region, and to the second creatine peak. Opstad et al. [26]

292

included glutathione at 2.9 ppm into their LCModel, and obtained a better fit. As discussed earlier, these authors reported that reduced glutathione was significantly greater in patients with meningiomas compared to those with astrocytomas. The authors of Ref. [27] found a resonance at 3.8 ppm at long TE, to which they did not give an assignment, but which was also characteristic of meningiomas, and was absent from other brain tumors. Further complicating the situation is that when patients are treated with mannitol, a resonance at 3.8 ppm will often appear, although in Ref. [27] only 2 of the 19 patients with meningiomas were receiving mannitol at the time of the MRS examination. Clearly exemplified here is the subjectivity of fitting procedures [3, 23], contrary to claims of objectivity [13], and how this can undermine diagnostic accuracy in neuro-oncology. Kaminogo et al. [15] reported that the NAA to choline ratio at short TE was of limited value in grading gliomas. They suggest that in gliomas the peak around 2 ppm could contain other metabolites besides NAA, namely lipids at 2.05 ppm, and glutamine/glutamate at 2.1 ppm. As reviewed in Chapter 8, the data concerning myoinositol are contradictory for distinguishing tumors from normal brain tissue and from non-neoplastic processes, as well as for brain tumor grading and histopathological typing. Most authors have assigned the peak at 3.56 ppm to myoinositol alone, while some have viewed this as a combined myoinositol-glycine peak. Glycine is an inhibitory neurotransmitter, which also has a proton signal at this position. After release, glycine is taken up by astrocytes, the site where myoinositol is also located [11, 28]. Whether better distinction between these two overlapping resonances of glycine and myoinositol would help clarify some of the above-mentioned diagnostic dilemmas remains to be determined. In vitro studies discussed in Chapter 8 provide some further insights into these issues. Auer et al. [29] note a major limitation of fitting procedures such as the LCModel in the presence of large amounts of mobile lipids. They state that prominent broad resonances at 0.9 and 1.3 ppm are not fully modelled by the baseline spline functions of the LCModel and, as such, lead to incorrect estimates of lactate and alanine. Inappropriate fit of the entire spectrum with substantial phasing errors can in some cases be the final, and wrong, result. These authors suggest some procedures to tackle this latter problem, to improve the accuracy of tumor grading using in vivo 1H MRS. They note, however, that limited spectral resolution and SNR, as well as a lack of prior knowledge about the specific lipid constituents of various disease states, presently limit the accuracy of quantifying lipid and macromolecular constituents. In their study using a long TE (272 ms), Kuznetsov et al. [30] chose to omit the peak at 0.9 ppm attributed to lipids, because they considered it could be biased by baseline fluctuations. Furthermore, they stated that

293

limitations in resolution prevented them from directly ascertaining the contribution of lipid versus lactate to the 1.3-ppm peak. In Ref. [27] mixed forms of lipid plus lactate doublet at 1.3 ppm were reported, using an even longer TE (288 ms). This peak was prominent in the patients with metastatic lesions. Dilemmas at both long and short TE concerning assignment and quantification, especially of overlapping resonances, were elaborated in the study by Fan et al. [31] in which clinical MRS findings from patients with high grade gliomas and solitary brain metastases were compared. Their observation that all 3 patients with recurrences had elevated lipids and glutamine-glutamate underscores the importance of proper assignment, as well as quantification. As emphasized by Bottomley [16], the areas of the peaks of most interest most often overlap {with} others (p. 3). The problem of overlapping metabolites was also shown to be of importance for prostate cancer diagnostics, as recently illustrated by Swanson et al. [32] who demonstrated that separating the overlapping polyamines from choline and creatine by in vivo 2D J-resolved MRS could be used to help detect the absence or presence of recurrent or residual prostate tumor (see Chapter 9). It should be noted that thus far, e.g. the LCModel has not been used for quantification of 2D MRS data. By contrast, the fast Pad transform has successfully quantified data from 2D MRS [33, 34].

17.5 Small number of observable compounds on MRS revealed with FFT

Using the FFT, only a few compounds (low molecular weight, high concentration) are observable on a clinical scanner, and these may not be the most critical for timely diagnosis of malignant processes. Limitations of the new neuro-chemistry in brain tumor diagnostics Danielsen and Ross [14] state that in vivo MRS has introduced a new neuro-chemistry, providing the clinician with the possibility of noninvasively accessing neuro-physiologic and neuro-chemical information. This, together with the exquisite detail of neuroanatomical imaging provided by MRI, offers an indispensable {tool} for state-of-the-art neuro-diagnosis (p. 23). However, the current Fourier-based applications of MRS and MRSI in brain tumor diagnostics yield a relatively small number of metabolites (low molecular weight, high concentration) that are observable on clinical scanners. In other words, in fact, we are provided with an

294

extremely limited view of normal and abnormal brain chemistry. As reviewed in Chapter 8, and summarized in Table 8.1, few of the metabolites identified by in vivo MRS unequivocally distinguish tumor from normal brain tissue. Moreover, none of these metabolites is pathognomonic for neoplastic tissue. On the contrary, each of them is entirely non-specific for brain malignancy, and the list of possible differential diagnoses is generally long and diverse. Perhaps the most notable of these is choline, which is the metabolite used most commonly for brain tumor diagnosis, but, as discussed, can also be elevated in infiltrative processes, sub-acute and chronic ischemia, encephalitis, demyelinization, Alzheimers disease, and epilepsy, inter alia. Indeed, as noted by Danielsen and Ross [14] despite the gold mine of clinical information that is to be gleaned from observation of alterations in the choline peak in MRS it is still the least well-understood in terms of its neuro-chemistry (p. 36). The choline peak seen on in vivo MRS is mainly comprised of phosphorylcholine and glycerophosphorylcholine, whereas phosphatidyl-choline, the major choline metabolite in the brain is MRinvisible [14]. As discussed further in Chapter 8, 2D correlated and in vitro MRS studies reveal that there is a much richer store of potentially informative metabolites for brain tumor diagnostics, than is currently extracted using in vivo 1D MRS and MRSI. Other informative metabolites for tumor diagnostics from 2D and in vitro MRS As reviewed in Part B of this book, an impressive number of metabolites that improve identification of cancerous tissue have been found using in vitro MRS. To recall a few examples:
Head and neck tumors Using in vitro proton 2D MRS, a number of amine acid resonances (alanine, glutathione, histidine, isoleucine, valine, and the shared cross peak for lysine/polyamine) were found in most or all squamous cell carcinomas of the head and neck, but not in normal tissue [35]. Breast cancer For breast cancer diagnostics using proton MRS, basing the diagnosis of malignancy upon a composite choline peak can be equivocal because, as noted, this can also be found in benign breast lesions and normal breast during lactation [36]. Use of longer TE to diminish overlap with the lipid signal further diminishes the number of compounds that can be visualized. The

295

composite choline peak used for breast cancer diagnosis in 1H MRS contains a number of water-soluble choline metabolites such as phosphocholine, glycerophosphocholine, betaine, analogous compounds containing the ethanolamine head group and taurine, as well as choline itself [5, 37]. Fibrocystic disease, fibroadenomas and tubular adenoma are some benign lesions that have also shown a composite choline peak. The reason why lactating women frequently also show a composite choline signal in their breast tissues is that milk is comprised predominantly of choline compounds such as phosphatidylcholine, phosphocholine etc., as well as free choline. Katz-Brull et al. [37] applied in vitro analysis using tracer kinetics and 13C and 31P MRS to examine the biochemical pathways underlying the high levels of water-soluble choline metabolites seen in breast cancer. They identified two non-intersecting pathways: phosphorylation and oxidation of choline, to be augmented with malignant transformation of mammary cells, with increased synthesis of PC and betaine. They also found suppression of choline-derived ether lipids. As reviewed, detailed comparisons among in vitro samples from breast cancer, fibroadenomas and normal breast tissue revealed that many other metabolites were more diagnostically informative than was total choline. Ovarian cancer As discussed in Chapter 10, in vivo MRS based upon choline, creatine, lactate and lipids did not provide clear distinction between ovarian malignancies and benign processes. However, an in vitro study [38] revealed that besides choline and lactate, ovarian cancer was distinguished from benign ovarian cysts by significantly higher concentrations of several amino acids including: isoleucine, valine, threonine, alanine, lysine, glutamine and methionine.

In the next chapter, we will explore how recent mathematical advances via the fast Pad transform applied to in vivo proton MRS might tap into this information and the potential implications for tumor diagnostics.

References
[1] J.K. Hennig, Fourier-based reconstruction techniques, MAGMA 15 (Suppl. 1), 36 (2002). [2] Dz. Belkic, Fast Pad transform for magnetic resonance imaging and computerized tomography, Nucl. Instr. Meth. Phys. Res. A 471, 165-169 (2001). [3] Dz. Belkic, Pad-based magnetic resonance spectroscopy, J. Comp. Meth. Sci. Eng. 3, 563733 (2003). [4] D.W. McRobbie, E.A. Moore, M.J. Graves, M.R. Prince, MRI from picture to proton, Cambridge University Press, Cambridge, 2003. [5] R. Katz-Brull, P.T. Lavin, R.E. Lenkinski, Clinical utility of proton magnetic resonance spectroscopy in characterizing breast lesions, J. Natl. Cancer Inst. 94, 1197-1203 (2002).

296

[6] K. Belkic, Magnetic resonance spectroscopic imaging in breast cancer detection: possibilities beyond the conventional theoretical framework for data analysis, Nucl. Instr. Meth. Phys. Res. A. 525, 313-321(2004). [7] K. Belkic, Current dilemmas and future perspectives for breast cancer screening with a Focus upon optimization of magnetic resonance spectroscopic imaging by advances in signal processing, Isr. Med. Assoc. J. 6, 610-618 (2004). [8] K. Belkic, The need for quantitative biomedical spectroscopic imaging through magnetic resonance in oncology beyond the conventional Fourier-based framework for signal processing, J. Comp. Meth. Sci. Engin 3, 535-561 (2003). [9] R. Kreis, 1H MR spectroscopy outside the brain, MAGMA 15 (Suppl 1), 2 (2002). [10] J.R. Griffiths, A.R. Tate, F.A. Howe, M. Stubbs, as part of the Multi-Institutional Group on MRS Application to Cancer, Magnetic resonance spectroscopy of cancerpracticalities of multicentre trials and early results in non-Hodgkins lymphoma, Eur. J. Cancer 38, 2085-2093 (2002). [11] L.A. Brando, R.C. Domingues, MR Spectroscopy of the Brain, Lippincott Williams & Wilkins, Philadelphia, Pennsylvania, 2004. [12] Y. Huang, P.J.G. Lisboa, W. El-Deredy, Tumor grading from magnetic resonance spectroscopy: a comparison of feature extraction with variable selection, Stat. Med. 22, 147-164 (2003). [13] S.W. Provencher, Estimation of metabolite concentrations from localized in vivo proton NMR spectra, Magn.Reson. Med. 30, 672-9 (1993). [14] E. R. Danielsen, B. Ross, Magnetic Resonance Spectroscopy Diagnosis of Neurological Diseases, Marcel Dekker, Inc., New York, 1999. [15] M. Kaminogo, H. Ishimaru, M. Morikawa, et al., Diagnostic potential of short echo time MR spectroscopy of gliomas with single-voxel and point-resolved spatially localised proton spectroscopy of brain, Neuroradiology 43, 353-363 (2001). [16] P.A. Bottomley, The trouble with spectroscopy papers, J. Magn. Reson. Imaging 2, 1-8 (1992). [17] D. Vigneron, A. Bollen, M. McDermott, et al., Three-dimensional magnetic resonance spectroscopic imaging of histologically confirmed brain tumors, Magn. Reson. Imaging 19, 89-101 (2001). [18] J.D. Rabinov, P.L. Lee, F.G. Barker, et al., In vivo 3-T MR spectroscopy in the distinction of recurrent glioma versus radiation effects: Initial experience, Radiology 225, 871-879 (2002). [19] B. Ciskowska-Lyson, L. Krolicki, A. Teska, et al., Proton magnetic resonance spectroscopy investigations in brain metabolic changes after first doses of chemotherapy, MAGMA 15 (Suppl.1), 149 (2002). [20] U.G. Mueller-Lisse, M.G. Swanson, D.B. Vigneron, et al., Time-dependent effects of hormone-deprivation therapy on prostate metabolism as detected by combined magnetic resonance imaging and 3D magnetic resonance spectroscopic imaging, Magn. Reson. Med. 46, 49-57 (2001). [21] C. Ars, Tumors and spectroscopy, MAGMA 15 (Suppl.1), 38 (2002). [22] Dz. Belkic, Non-Fourier based reconstruction techniques, MAGMA 15 (Suppl.1), 36-37 (2002). [23] Dz. Belkic, Fast Pad Transform (FPT) as opposed to conventional fitting for signal and image processing; International Conference on Imaging Techniques in Subatomic Physics, Astrophysics, Medicine, Biology and Industry-Book of Abstracts, Stockholm, June 24-27, 2003, p. 103.

297

[24] L. Vanhamme, A. van den Boogart, S. van Huffel, Improved method for accurate and efficient quantification of MRS data with use of prior knowledge, J. Magn. Reson. 29, 35-43 (1997). [25] A. van den Boogaart, Quantitative data analysis of in vivo MRS data sets, Magn. Reson. Chem. 35, S146-S152 (1997). [26] K.S. Opstad, S.W. Provencher, B.A. Bell, et al., Detection of elevated glutathione in meningiomas by quantitative in vivo 1H MRS, Magn. Reson. Med. 49, 632-637 (2003). [27] Y-D. Cho, G-H. Choi, S-P. Lee, J-K. Kim, 1H-MRS metabolic patterns for distinguishing between meningiomas and other brain tumors, Magn. Reson. Imaging 21, 663-672 (2003). [28] E.J. Novotny, R.K. Fulbright, P.L. Pearl, et al. Magnetic resonance spectroscopy of neurotransmitters in human brain, Ann. Neurol. 54, S25-S31 (2003). [29] D.P. Auer, C. Gssl, T. Schirmer, M. Czisch, Improved analysis of 1H-MR spectra in the presence of mobile lipids, Magn. Reson. Med. 46, 615-618 (2001). [30] Y.E. Kuznetsov, Z. Caramanos, S.B. Antel, et al., Proton magnetic resonance spectroscopic imaging can predict length of survival in patients with supratentorial gliomas, Neurosurgery 53, 565-576 (2003). [31] G. Fan, B. Sun, Z. Wu, et al., In vivo single-voxel proton MR spectroscopy in the differentiation of high-grade gliomas and solitary metastases, Clin. Radiol. 59, 77-85 (2004). [32] M.G. Swanson, D.B. Vigneron, T-K.C. Tran et al., Single-voxel over-sampled J-resolved spectroscopy of in vivo human prostate tissue, Magn. Reson. Medicine 45, 973-980 (2001). [33] Dz. Belkic, High-resolution parametric estimation of two-dimensional magnetic resonance spectroscopy, European Society for Magnetic Resonance in Medicine and Biology 2003, Rotterdam, 365 (CD). [34] N. Binesh, S. Han, A. Kumar, K. Yue, M.A. Thomas, First clinical applications of localized 2D COSY, MAGMA 15 (Suppl. 1), 128-129 (2002). [35] S.K. Mukherji, S. Schiro, M. Castillo et al., Proton MR spectroscopy of squamous cell carcinoma of the extracranial head and neck: in vitro and in vivo studies, Am. J. Neuroradiol. 18, 1057-1072 (1997). [36] K.A. Kvistad, I.J. Bakken, I.S. Gribbestad, et al., Characterization of neoplastic and normal human breast tissues with in vivo (1) H MR spectroscopy, J. Magn. Reson. Imaging 10, 159-164 (1999). [37] R. Katz-Brull, D. Seger, D. Rivenson-Segal, E. Rushkin, H. Degani, Metabolic markers of breast cancer: enhanced choline metabolism and reduced choline-ether-phospholipid synthesis, Cancer Res. 62, 1966-1970 (2002). [38] E.A. Boss, S.H. Moolenaar, L.F.A.G. Massuger, H. Boonstra, U.F.H. Engelke, J.G.N. de Jong, R.A. Wevers, High-resolution proton nuclear magnetic resonance spectroscopy of ovarian cyst fluid, NMR Biomed. 13, 297-305 (2000).

299

Chapter 18

Mathematical Advances in Spectral Analysis: Relevance for MRS/MRSI Cancer Diagnostics

_______________________________________________________________________________

It has been conclusively shown in a series of recent papers [1-13] that the fast Pad transform can overcome many of the limitations of the FFT that are relevant to in vivo MRS and MRSI. Recall that the FPT is a non-linear rational approximation of two frequency dependent polynomials P/Q to the Maclaurin expansion whose coefficients are the raw time signal points {cn}. For a brief mathematical background of the FPT, presented in terms that can be understood by a broader audience, including those from the medical and biological sciences, the reader is referred to Chapter 4 of this book.

18.1 The fast Pad transform has rapid, stable convergence

In Chapter 17, we focused upon the limitations of the Fast Fourier Transform. On the other hand, the FFT has some highly desirable properties. One of these, which is perhaps the main reason for its general popularity, particularly in signal analysis for clinical purposes, is that the FFT shows steady convergence with increasing signal length N, and reasonable looking spectra for truncated signals. This steady convergence means that there are no major troublesome surprises as the values of N are systematically augmented. In sharp contrast, nearly all of the parametric estimators show marked instability as a function of N. This is typically manifested in dramatic oscillations (e.g. spikes and other artificial spectral structures) that appear prior to convergence. Where to stop the calculations under wild convergence patterns is also a matter of concern. Needless to say,

300

such spurious findings are totally anathema to the clinical demands for reliable spectral information. Unlike a number of the other available parametric estimators, the FPT demonstrates extremely stable convergence with increasing signal length, as recently shown in Refs. [6-8]. Throughout the entire convergence test, starting from small fractions up to the complete signal length, the FPT shows no spurious findings whatsoever, i.e. no spikes, no unphysical resonances, no aliasing, no Gibbs oscillations or any other artefacts that might stem from signal truncation. One of key advantages of the FPT relative to the FFT is that convergence is not only stable, but also rapid. This means that even at short signal lengths, the FPT is still capable of determining the true concentrations of the main metabolites that remain undetected by the FFT.

18.2

The FPT improves resolution and SNR

Extrapolation and interpolation by the FPT Recall that the resolution of the FFT is predetermined by the total acquisition time T. Whenever the signal length N is finite, truncation artefacts appear as aliasing, Gibbs phenomena and other shape distortions. At short signal lengths, these truncation artefacts can be very troublesome. Attempts to improve the appearance of the Fourier shape spectra are made by replacing any extension of the signal by zeros (cn = 0, n>N) or using the signals periodic extension (cn+N =cn). However, neither of these strategies adds any new information. Thus, particularly for short signal lengths, there is a great need for estimators that can extrapolate with fidelity beyond the acquisition time T. As reviewed in Chapter 4, the FPT achieves this extrapolation because of the implicit infinite sum contained in the inverse polynomial Q 1 since, of course, P Q = PQ 1 . From here, although both P and

Q are finite sums, the inverse Q 1 is a series, i.e. an infinite sum, such that the product PQ 1 is also an infinite sum. Hence, extrapolation. In
other words, an originally finite sum is now mapped into an infinite sum by using only a finite number of available signal points. Also noted in Chapter 4, the spectrum in the FPT does not use the fixed Fourier mesh in the frequency domain, and therefore is not predetermined by the total acquisition time T. Rather, the resolution of the FPT is the average distance among the adjacent frequencies in a given interval of the full Nyquist range. This average distance is almost

301

always smaller than the Fourier minimal distance. Hence improved interpolation and better resolution. The conundrum between increasing acquisition time T for improved resolution and increasing noise is thereby obviated by the FPT. This is especially important for accurate detection of short-lived metabolites. Improvements in resolution do not violate the Heisenberg uncertainty principle The often-raised argument that improvements beyond the Fourier resolution would violate basic principles of physics (i.e. the Heisenberg uncertainty principle) has also been thoroughly refuted with respect to the FPT. This is because there is only one physical quantity which is measured, i.e. the time signal c (t ) , whereas the spectrum in MRS is generated theoretically via signal processing (recall the discussion of conjugate variables in Chapter 4). Since the signal processing, which, at any rate, takes place after the measurement, cannot possibly destroy the measurement itself, it follows that the resolution limitation is only due to the weaknesses of linear algebra, such as in the FFT [2]. Therefore, the Fourier resolution can be improved, as is done by all parametric estimators. In other words, these estimators may well go beyond the Fourier uncertainty principle, which has nothing to do with the Heisenberg uncertainty principle. The FPT is a non-linear mappingpermitting noise suppression As discussed in Chapter 17, the FFT has poor SNR due not only to the need for long acquisition times, but also because it is a linear mapping, since its transformation coefficients or weights are independent of the time signal points. The FPT, in contrast, is a non-linear mapping, such that its coefficients are dependent upon the time signal points. As opposed to the FFT whose linearity preserves noise from the time signal, the non-linearity of the FPT permits noise suppression. The polynomial quotient from the FPT can be recognized as the more familiar auto-regressive moving average (ARMA) process [2]. Moreover, the FFT has a linear convergence ( 1/N) with increased signal length N, whereas the convergence of the FPT is at least quadratic ( 1/N2 ), if not better [2]. Specific advantages of improved resolution and SNR for MR-based cancer diagnostics We have noted that a number of the problems with current applications of MRS and MRSI in tumor diagnostics are related to resolution and

302

SNR issues. These difficulties include detection of very small tumors, accurate tumor grading, spatial resolution especially for single-voxel studies, as well as problems with short TEs, inter alia. As discussed in Chapter 4, the inverse Fourier transform can be used to obtain spatial information. The FPT can be applied here to increase resolution since it is not bound to the Fourier grid. In MRI the key is simultaneous brightness and sharpness. However, the Fourier demand for an increased acquisition time inevitably leads to decreased brightness and since this is related to noise, sharpness is compromised. Due to its extrapolation feature, the FPT allows sampling at shorter acquisition time and this enhances both sharpness and brightness, as well as diminishing the Gibbs phenomenon [1, 14, 15].

18.3 The FPT determines the exact number of metabolites

As reviewed in Chapter 4, the FPT provides precise numerical quantification without any free or adjustable parameters, i.e. without fitting, because it computes the spectra through the unique ratio of two frequency dependent polynomials [2]. A key advantage of the FPT is its unequivocal specification of the exact number K of metabolites from the encoded time signals. This number K is, in fact, contained in the encoded signal, and is extracted by the FPT applied directly to the raw time signal. This signal is stored as a data matrix (also called a Hankel matrix), HK (cn). The condition that is sought is that the signal has K metabolites, which occurs when the following requirements are simultaneously fulfilled:

HK+1 (cn) = 0,

HK (cn) 0.

(18.1)

In practice, the Hankel determinants of increasing order {Hm(cn)} (m = 1, 2, N/2) are computed. The first m=m at which (18.1) is satisfied, gives the total number of metabolites, K=m. This procedure of obtaining the number of metabolites K exactly from the raw signal points was initially suggested particularly for MRS in Ref. [2, 4, 6].

303

18.4 The FPT treats both Lorentzian and nonLorentzian spectra on the same footing
We also noted in Chapter 4 that the FPT can treat both Lorentzian (nondegenerate) and non-Lorentzian (degenerate) spectra on the same footing. This is a distinct advantage of the FPT relative to, e.g. the HLSVD algorithm of the MRUI [16], which is limited to Lorentzians only. The FPT obtains the amplitude of each metabolite separately from an analytical expression, which depends only upon the frequency of the metabolite being quantified. This is distinctly advantageous compared to other methods such as the HLSVD, which uses all found (genuine as well as spurious) frequencies for each amplitude, such that errors in some found frequencies can seriously undermine the accuracy of the estimated amplitude [4].

18.5 The FPT can unambiguously identify overlapping resonances

The FPT can identify spurious resonances by analytical methods; this is followed by a well-defined procedure for regularization, via the socalled constrained root reflection with preservation of the magnitude or power spectra. Overlapping or hidden metabolites, including those that may be disguised in noise, are retrieved with fidelity [2]. In Chapter 17, we noted that fitting and assignment became particularly troublesome for overlapping or closely lying metabolites. It was also pointed out that the areas of the peaks of most clinical interest for oncology very often overlap with others.

18.6 The FPT accurately estimates concentrations

In contrast to the FFT, which is a non-parametric method supplying only a shape spectrum, the parametric FPT yields precise estimates of all the relevant peak parameters (position, heights, widths, phases) of every true metabolite. (See Chapter 4 for a brief discussion of how this is achieved mathematically, and references cited therein for a detailed review). The FPT provides an exact numerical quantification without any free or adjustable parameters, i.e. without fitting [2], because it computes that spectra through the unique ratio of two frequency dependent polynomials P Q , by using only the originally encoded time signal. This feature could offer an important advantage in improving the diagnostic accuracy of MRS. Namely, rather than relying upon a

304

determination of presence or absence of a given metabolite, e.g. the actual peak height and full width at half-maximum could be used to specify the normal range versus values seen in malignancy.

18.7 An illustration of the performance of the FPT for a clinical MRS signal
We now present an illustration of the performance of the FPT for a clinical MRS signal obtained from a recording of the brain of a healthy volunteer. We use the measured time domain data (FID) acquired at the static magnetic field strength of 4T. These data of full signal length N = 2048 encoded by the group at the Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, USA [17] have been kindly made available to us. In Figure 18.1, we present absorption spectra at three signal lengths, comparing the FFT (left column) and the FPT (right column). At the top of Figure 18.1, the most dramatic difference between the FFT and FPT is seen at the shortest signal length (N/16 = 128). Here, the FFT essentially presents no meaningful spectroscopic information. In contrast, with the FPT, at (N/16 = 128) nearly 90% of the NAA concentration is predicted by the peak at around 2.0 ppm. In the middle panel it is seen that at N/4 = 512, the FFT has still not predicted even 70% of the NAA concentration at 2.0 ppm, and the ratio between creatine and choline (3.0 and 3.3 ppm) appears to be nearly equal, and thus wrong. In contrast, with the FPT at N/4 = 512, these three major peaks are now practically identical to those at full signal length. At half signal length (N/2 = 1024) at the bottom panel, the FFT has still not demonstrated the accurate ratio between creatine and choline at 3.0 and 3.3 ppm, respectively; these two metabolites are still incorrectly appearing as almost equal. Moreover, the triplet of glutamine and glutamate near 2.3 ppm can be discerned at half signal length only by the FPT, and not by the FFT. By contrast, it is seen that at half signal length (N/2 = 1024) the FPT resolves with fidelity more than twenty metabolites, in which all peak parameters are accurately extracted, including the overlapping resonances. Furthermore, while the FFT demands the total signal length (N = 2048) to fully resolve all the metabolites, the difference between the two FPT spectra at N = 1024 and N = 2048 is buried entirely in the background noise [18]. In other words, the FPT spectra at half-signal length can be treated as fully converged. Most importantly, as is clear from Figure 18.1, the FPT produces no spurious metabolites or other spectral artefacts in the process of converging in a strikingly steady fashion as a function of the increased signal length. Moreover, it is obvious that the FPT exhibits a

305

much faster convergence rate than that in the FFT (for further illustrations see Ref. [6, 7, 18].
Figure 18.1 Fourier and Pad absorption spectra computed using the time signal (divided by 10 000) at 3 truncated signal lengths (N/16 =1 28, N/4 = 512, N/2 = 1024), where the full signal length is N = 2048, as encoded in Ref. [17] at 4T from occipital grey matter of a health volunteer. From Ref. [11], with
permission.
FOURIER 2.5
Absorption, (F) [a.u.]

PADE 2.5
Absorption, (P/Q) [a.u.]

2 1.5

FFT N/16 = 128

2 1.5 1

FPT N/16 = 128

1 0.5 0

0.5 0

0.5

2 3 4 Chemical shift (ppm)

0.5

2 3 4 Chemical shift (ppm)

2.5
Absorption, (F) [a.u.]

2.5
Absorption, (P/Q) [a.u.]

2 1.5 1

FFT N/4 = 512

2 1.5 1

FPT N/4 = 512

0.5 0

0.5 0

0.5

2 3 4 Chemical shift (ppm)

0.5

2 3 4 Chemical shift (ppm)

2.5
Absorption, (F) [a.u.]

2.5
Absorption, (P/Q) [a.u.]

2 1.5 1

FFT N/2 = 1024

2 1.5 1

FPT N/2 = 1024

0.5 0 0.5

0.5 0 0.5

2 3 4 Chemical shift (ppm)

2 3 4 Chemical shift (ppm)

The abscissa represents chemical shift in dimensionless units (ppm). The ordinates are intensities in arbitrary units (a.u.). The symbols (F) and (P/Q) denote the real part of the complex Fourier and Pad spectrum F and P/Q, respectively.

306

18.8 Validity of the FPTError analysis

The present example was obtained from a signal encoded at 4T, with excellent SNR. Signals encoded at lower field, e.g. 1.5T, as in most clinical scanners, have poorer SNR compared to the one above. In such cases, the gold standard of an excellent shape spectrum from the FFT is unavailable. The FPT has been applied in the latter cases as well, with validity assessment achieved by internal cross validation using two (+) (-) variants of the FPT. These are: FPT and FPT that have their convergence regions inside (|z| < 1) and outside (|z| > 1) the unit circle, respectively [7, 9]. Once convergence has been achieved with increased signal length, the resulting fundamental frequencies from these two variants become infinitesimally close to the points on the unit circle (|z| = 1) from both sides. This joint set of complex frequencies {k} and corresponding complex amplitudes {dk} constitute the sought gold standard. The results from the FPT presented in Figure 18.1 were (-) performed in the FPT . Extensive error analysis has been performed by computing the residual or error spectra, defined as the difference between the spectra in the (+) (-) FPT and FPT at a given signal length. Such spectra are used to answer the question: where does one stop in the convergence process within the Pad method? This question is answered in two parts. Firstly, one stops when the residual spectra as a function of the signal length become indistinguishable from the background noise, e.g. the root mean square (rms). Secondly, one stops when constancy is reached in the values of all four parameters (position, height, width and phase) of each genuine, physical resonance, for varying signal length. Thus, these two conditions: (a) indistinguishability of the residual spectra from the background noise, and (b) stabilization of spectral parameters, fulfil the requisites of error analysis of proven validity [10, 19].

18.9 Appropriateness of the FPT for tumor diagnostics within magnetic resonance
We have reviewed extensively how dilemmas surrounding assignment, the non-uniqueness of fitting and quantification represent major obstacles for clinical MRS and MRSI with respect to tumor identification, histological classification, tumor grading, assessment of response to therapy and early detection of tumor recurrence.

307

In short, the fast Pad Transform fulfills the most stringent requirements for tumor diagnostics within magnetic resonance:

Markedly enhanced resolution and SNR compared to the FFT, As a parametric method provides precise numerical data for all peak parameters (position, height, width and phase) for every true metabolite, Specifies the exact number of metabolites from the encoded time data (FID) and can identify unambiguously overlapping metabolite, as well as metabolites present in low concentrations, Obtains the amplitude of each metabolite separately from an analytical expression dependent only upon the metabolites frequency, Handles both Lorentzian and spectra on the same footing, non-Lorentzian

Strikingly robust and stable convergence for varying fractions of the full signal length, yielding reasonable concentrations of the main metabolites even for severely truncated time signals, Rigorous validation and error analysis have been performed, Straightforward and standard programming with absolutely minimal computer resources and time, Extends readily to multidimensional signal/image processing.

308

References
[1] Dz. Belkic, Fast Pad transform for magnetic resonance imaging and computerized tomography, Nucl. Instr. Meth. Phys. Res. A. 471, 165-169 (2001). [2] Dz. Belkic, Exact analytical expressions for any Lorentzian spectrum in the Fast Pad transform (FPT), J. Com. Meth. Sci. Eng. 3, 109-186 (2003). [3] Dz. Belkic, Non-Fourier based reconstruction techniques, MAGMA 15 (Suppl.1), 36-37 (2002). [4] Dz. Belkic, Pad-based Magnetic Resonance Spectroscopy (MRS), J. Com. Meth. Sci. Eng. 3, 563-733 (2003). [5] Dz. Belkic, Fast Pad Transform (FPT) as opposed to conventional fitting for signal and image processing; Int. Conf. Imaging Book of Abstracts, Stockholm, June 24-27, 2003, p. 103. [6] Dz. Belkic, Strikingly stable convergence of the fast Pad transform: Applications to Magnetic Resonance Spectroscopy, J. Com. Meth. Sci. Eng. 3, 299-382 (2003). [7] Dz. Belkic, Strikingly stable convergence of the Fast Pad transform (FPT) for high-resolution parametric and non-parametric signal processing of Lorentzian and non-Lorentzian spectra, Nucl. Instr. Meth. Phys. Res. A. 525, 366-371 (2004). [8] Dz. Belkic, Quantum Mechanical Signal Processing and Spectral Analysis, Institute of Physics Publishers, Bristol, 2004. [9] Dz. Belkic, Analytical continuation by numerical means in spectral analysis using the fast Pad transform (FPT), Nucl. Instr. Meth. Phys. Res. A. 525, 372-378 (2004). [10] Dz. Belkic, Error analysis through residual frequency spectra in the fast Pad transform (FPT), Nucl. Instr. Meth. Phys. Res. A. 525, 379-386 (2004). [11] K. Belkic, The need for quantitative biomedical spectroscopy imaging through magnetic resonance in oncology beyond the conventional Fourier-based signal processing, J. Com. Meth. Sci. Eng. 3, 535-561 (2003). [12] K. Belkic, Current dilemmas and future perspectives for breast cancer screening with a focus upon optimization of magnetic resonance spectroscopic imaging by advances in signal processing, Isr. Med. Assoc. J. 6, 610-618 (2004). [13] K. Belkic, Magnetic resonance spectroscopic imaging in breast cancer detection: possibilities beyond the conventional theoretical framework for data analysis, Nucl. Instr. Meth. Phys. Res. A. 525, 313-321 (2004). [14] Dz. Belkic, K. Belkic, High resolution magnetic resonance imaging (MRI), IEEE Medical Imaging Conference (MIC), 2003, Abstract Number 1971, Portland, October 2003. [15] M.F. Callaghan, D.J. Larkman, M.C. Wiltshire, J.V. Hajnal, Edge detection at low resolution using Pad approximants, Proc. Intl. Soc. Mag. Reson. Med. 11, 754 (2004). [16] A. van den Boogaart, Quantitative data analysis of in vivo MRS data sets, Magn. Reson. Chem. 35, S146-S152 (1997). [17] I. Tk, P. Andersen, G. Adriany, et al., In vivo 1H NMR spectroscopy of the human brain at 7T, Magn. Reson. Med. 46, 451-456 (2001). [18] K. Belkic, Dz. Belkic, Spectroscopic imaging through magnetic resonance for brain tumor diagnostics, J. Com. Meth. Sci. Eng. 4, 132-180 (2004). [19] Dz. Belkic, K. Belkic, Error analysis and residual spectra in the Fast Pad transform (FPT) for magnetic resonance spectroscopy (MRS), ESMRB, 2004, Copenhagen, #183 CD.

309

Chapter 19

Next Needed Steps: Pad-Based Optimization of MRS and MRSI in Cancer Diagnostics
_______________________________________________________________________________

Thus far, the advances in signal processing offered by the fast Pad transform have been applied to assess MR spectra encoded from healthy individuals. As we saw in Chapter 18, based upon a series of recent papers [1-14] the advantages of the FPT over the conventional Fourier-based framework for processing MRS signals are clear. Namely, it was unequivocally demonstrated that the FPT can be used to extract invaluable and unique quantitative information about brain metabolites. The FPT is currently being applied to provide norms, based upon quantification of those peak parameters in healthy volunteers. This has begun with proton-MRS for the various brain regions, to be continued for organs outside the brain, whenever possible using both 1H MRS and 31P MRS [11]. While the establishment of norms/standards is, of course, essential, we concluded in the last chapter, that the FPT fulfills the most stringent requirements for analysis of MRS signals as needed for tumor diagnostics. Therefore, the next, urgently needed step is to apply the FPT to process in vivo MRS signals encoded from patients with cancers. Multivariate analysis would then be performed to determine the best set of metabolic predictors of malignancy, on a tumor-specific basis, with quantification. Another vital component will be to perform Pad-based analysis of MRS data from non-malignant lesions that have heretofore presented differential diagnostic dilemmas, most notably benign tumors, infectious or inflammatory lesions, as reviewed in detail in Part B of this book.

310

Insofar as the FPT indeed proves to provide the predicted improvement in diagnostic accuracy of clinical MRS for cancer detection, then Padbased MRS could then be used more widely in clinical oncology. Heretofore, we have discussed applications of Pad-based signal analysis for MRS. The clear advantages of MRSI in providing volumetric coverage for tumor diagnostics have been shown repeatedly in Part B of this book. Since the time signals from MRSI are precisely of the same nature as those from MRS, the FPT can also be applied with equal success to MRSI, and this has been accomplished [15]. Another application of the FPT is for in vivo 2D MRS [16]. Here, the cross-correlation diagrams showed marked improvement in resolution and shortening of the signal length relative to 2D FFT. In 2D MRS one of the time axes is usually long, but the other must be short to maintain a reasonable scanning time. This is where estimators that can extract information from short time signals become indispensable. The FPT has been shown to fulfill this task, which is virtually unfeasible by attempts to extend the one-dimensional fitting algorithms (LCModel, AMARES, etc.) to two-dimensions. Indeed, the possibilities of 2D J resolved MRS to extract vital information for tumor detection have been underscored [17]. With the aid of Pad-based data analysis, these advantages of 2D MRS might be better utilized for cancer diagnostics.

19.1 View to early detection / screening

19.1.1 Challenges entailed in cancer screening As emphasized by Rimer et al. [18], while the aim of cancer screening is very practical, i.e. the detection of cancer at an early and treatable stage, the reality is quite complex (p. 627). Most challenging is to ensure that outcome is truly improved, and that harm does not occur. This, of course, is the basic principle of all medicine: Primam non nocere--above all, do no harm. However, since screening is performed on asymptomatic persons, a special set of thorny problems must be addressed [19]. Overall, the positive aims of screening for cancer include [18]:
Improved prognosis for detected by screening, those in whom cancer is

Need for a less radical treatment in those in whom cancer is detected at an earlier stage,

311

Confidence that a negative cancer is truly not present,

test

means

that

the

Lowered costs from less radical treatment.

In order for cancer screening to be considered, the above-outlined benefits should be clearly attainable. Moreover, the negative consequences should not override the benefits. The following are some of the potential important adverse effects of cancer screening [18, 19]:
False positive findings, and the attendant medical, psychological and economic consequences, False negative findings, and the clinical and other consequences, Potential carcinogenic effects of the screening procedure, as well as other possible negative aspects of the procedure including discomfort, inconvenience, and cost, Over-diagnosis--detection and treatment of a slowgrowing cancer that would never have caused medical problems, side effects of treatment plus the psychosocial impact of receiving a diagnosis of cancer.

19.1.2 Optimization of MRSI for early detection of cancers As yet, only under very limited circumstances have MRS and MRSI been used for early detection of the most prevalent cancers. Problems of resolution and SNR have been important obstacles for identifying small tumors. The lack of specificity as well as quantification issues, are major drawbacks of the currently used metabolic indicators of malignancymost notably choline. In vitro studies have shown that short-lived metabolites as well as those which overlap can provide insights for diagnosing cancers. As reviewed in Chapter 17, these limitations in the current applications of MRS and MRSI for early cancer detection are closely linked to reliance upon the conventional Fourier-based approach to data analysis. We have demonstrated in Chapter 18, based upon Refs. [1-15], that the FPT fulfils the most stringent requirements for tumor diagnostics, and in fact is the method of choice for processing MRS signals [15].

312

Now the strategic question arises: where might these advantages of the FPT yield the greatest potential benefits with respect to early diagnosis and screening? Pad-based MRSI for breast cancer screening We have explored this issue in detail with respect to breast cancer [11, 12, 14], and concluded that this would be one very important avenue to pursue, for a number of reasons that we will briefly outline here. Firstly, is the clear survival benefit achieved by early detection [20], coupled with the fact that breast cancer represents a major public health problem, being the leading cause of cancer-related deaths among women [21]. Secondly, mammography, the currently recommended screening method, has a number of major drawbacks including the potential carcinogenic effects of radiation exposure to the breast which is a radiosensitive organ, sub-optimal sensitivity such that one cannot confidently say that breast cancer has been ruled out by a normal mammogram, and very poor specificity, such that up to 80-90% of breast biopsies show benign findings [22]. The deficiencies of mammography are particularly troublesome with regard to younger women at high risk for breast cancer, the very group who would require greater scrutiny. As noted, MRI is now being quite widely tested among such women, with the benefits including no exposure to ionizing radiation and improved sensitivity, compared to mammography [23, 24]. However, MRI shares with mammography the problem of poor specificity [25]. In a meta-analysis of the few studies to date using 1HMRS, a substantially improved, albeit less than ideal, specificity was achieved relative to MRI with respect to breast cancer diagnosis. Moreover, the technical requirements are quite minimal, such that routine addition of MRS to an MRI exam of the breast is considered entirely feasible [25]. The limitations to the current applications in vivo MRS with respect to breast cancer diagnostics are very closely related to the reliance upon conventional Fourier based techniques for signal processing [12, 14]. These limitations include: poor resolution and SNR such that very small tumors are not detected, the need for lipid suppression which requires longer TE, further compromising SNR, as well as reducing the number of visible resonances. As a consequence, qualitative assessment of a single metabolite (total choline) is used to make the diagnosis of breast cancer. This leads to substantial non-differential misclassification, especially since the presence or absence of choline is not a unique finding for breast cancer, but can also be seen with benign breast disease, as well as during lactation, inter alia. Moreover, risk of breast cancer is increased in the former, and can coexist with the latter,

313

emphasizing the vital importance of characterizing the metabolic profile of each of these entities as precisely as possible, with quantification. As summarized in Chapter 16, detailed paired and logistic regression analysis [12, 14] confirmed the conclusion of the original authors [26] that a very rich window of information is provided by in vitro 1H MRS analysis of metabolite concentrations in malignant versus noncancerous breast tissue. This rich source of information could indeed be tapped with Pad-based signal processing of MRS and MRSI signals from the breast [12, 14]. We believe that this possibility should be urgently explored with the aim of optimizing early breast cancer detection, especially for younger women at high risk. Surveillance of high risk groups for specific cancers Certainly, there are many other areas in oncology where Pad-based MRSI should also be explored for screening and early detection, especially among high risk groups. Perhaps one of the most salient examples is for women at high risk of developing ovarian cancer due to the presence of BRCA 1 or BRCA 2 mutations. These mutations confer a 15 to 65% lifetime ovarian cancer risk (see Chapter 10). In light of the lack of effective screening and the poor prognosis of advanced disease, bilateral oophorectomy has been recommended as a prophylactic strategy, notwithstanding the drastic nature of such an intervention [27]. As mentioned, technical challenges especially related to motion artefacts as well as the small size of the organ, have rendered in vivo MRS very difficult for evaluation of the ovary, and very few such studies have been published. On the other hand, in vitro investigations [28-31] did reveal a panoply of MR spectroscopic findings that distinguish malignant from benign ovarian lesions. As suggested by Massouger et al. [31] insofar as the current obstacles hindering the acquisition of high quality spectra could be surmounted, in vivo 1H MRS could become the method of choice for evaluating such lesions. In Chapter 9 we noted that MRSI has been suggested as a potentially helpful modality for assessing patients with a high index of suspicion for prostate cancer, despite an initial negative biopsy [32]. Once again, in vitro studies illustrate that many more spectral characteristics (e.g. levels of spermine, taurine, myoinositol and glutamate, inter alia) help distinguish cancer of the prostate from benign or normal tissue, besides the choline to citrate ratio upon which in vivo MRSI diagnosis of prostate cancer is most heavily based. Given the fact that MRSI is

314

already widely applied in prostate cancer diagnostics, Pad-based optimization would be very feasible, and is another recommended priority. This is particularly important because of the current dilemmas surrounding prostate cancer screening, as well as the public health importance of this malignancy.

References
[1] Dz. Belkic, P.A Dando, J. Main, et al. Three novel high-resolution nonlinear methods for fast signal processing, J. Chem. Phys. 113, 6542-6556 (2000). [2] Dz. Belkic, Fast Pad transform for magnetic resonance imaging and computerized tomography, Nucl. Instr. Meth. Phys. Res. A. 471, 165-169 (2001). [3] Dz. Belkic, Non-Fourier based reconstruction techniques, MAGMA 15 (Suppl.1), 36-37 (2002). [4] Dz. Belkic, Exact analytical expressions for any Lorentzian spectrum in the Fast Pad transform (FPT), J. Com. Meth. Sci. Eng. 3, 109-186 (2003). [5] Dz. Belkic, Fast Pad Transform (FPT) as opposed to conventional fitting for signal and image processing; International Conference on Imaging Techniques in Subatomic Physics, Astrophysics, Medicine, Biology and Industry-Book of Abstracts, Stockholm, June 24-27, 2003, p. 103. [6] Dz. Belkic, Strikingly stable convergence of the fast Pad transform: Applications to Magnetic Resonance Spectroscopy, J. Com. Meth. Sci. Eng. 3, 299-382 (2003). [7] Dz. Belkic, Pad-based Magnetic Resonance Spectroscopy (MRS), J. Com. Meth. Sci. Eng. 3, 563-733 (2003). [8] Dz. Belkic, Strikingly stable convergence of the Fast Pad transform (FPT) for high-resolution parametric and non-parametric signal processing of Lorentzian and non-Lorentzian spectra, Nucl. Instr. Meth. Phys. Res. A. 525, 366-371 (2004). [9] Dz. Belkic, Analytical continuation by numerical means in spectral analysis using the fast Pad transform (FPT), Nucl. Instr. Meth. Phys. Res. A. 525, 372-378 (2004). [10] Dz. Belkic, Error analysis through residual frequency spectra in the fast Pad transform (FPT), Nucl. Instr. Meth. Phys. Res. A. 525, 379-386 (2004). [11] K. Belkic, The need for quantitative biomedical spectroscopy imaging through magnetic resonance in oncology beyond the conventional Fourier-based signal processing, J. Com. Meth. Sci. Eng. 3, 535-561 (2003). [12] K. Belkic, Magnetic resonance spectroscopic imaging in breast cancer detection: possibilities beyond the conventional theoretical framework for data analysis, Nucl. Instr. Meth. Phys. Res. A. 525, 313-321 (2004). [13] K. Belkic and Dz. Belkic, Spectroscopic imaging through magnetic resonance for brain tumor diagnostics: Recent achievements, dilemmas and potential solutions via advances in signal processing, J. Com. Meth. Sci. Eng. 4, 132-180 (2004). [14] K. Belkic, Current dilemmas and future perspectives for breast cancer screening with a focus upon optimization of magnetic resonance spectroscopic imaging by advances in signal processing, Isr. Med. Assoc. J. 6, 610-618 (2004). [15] Dz. Belkic, K. Belkic, Extremely stable and rapid convergence of the Fast Pad Transform (FPT) for high resolution processing of magnetic resonance spectra (MRS), Eur. Soc. Magn. Reson. Med. Biol., Copenhagen, 2004 # 343 (CD).

315

[16] Dz. Belkic, High resolution parametric estimation of two-dimensional magnetic resonance spectroscopy, Eur. Soc. Magn. Reson. Med. Biol., Rotterdam, 2003, # 365 (CD). [17] M.A. Thomas, L.N. Ryner, M.P. Mehta, et al., Localized 2D J-resolved 1H MR spectroscopy of human brain tumors in vivo, J. Magn. Reson. Imaging 6, 453-459 (1996). [18] B.K. Rimer, H. Schildkraut, R.A. Hiatt, Cancer screening, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 627-640. [19] O.W. Brawley, B.S. Kramer, Prevention and early detection of cancer, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 497-502. [20] R. A. Smith, D. Saslow, K.A. Sawyer, et al. American Cancer Society guidelines for breast cancer screening: update 2003. CA. Cancer J. Clin. 53, 141-169 (2003). [21] S. Hankinson, D. Hunter, Breast cancer, In: Adami H-O, Hunter D, Trichopoulos D. Textbook of Cancer Epidemiology, New York, NY: Oxford University Press, 2002, p. 301-339. [22] M.E. Lippman, Breast cancer, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001,p. 571-578. [23] E.A. Morris, Breast cancer imaging with MRI, Radiol. Clin. N. Am. 40, 443-446 (2002). [24] S.G. Orel, M.D. Schnall, MR imaging of the breast for the detection, diagnosis, and staging of breast cancer, Radiology 220, 13-30 (2001). [25] R. Katz-Brull, P.T. Lavin, R.E. Lenkinski, Clinical utility of proton magnetic resonance spectroscopy in characterizing breast lesions, J. Natl. Cancer Inst. 94, 1197-1203 (2002). [26] I. S. Gribbestad, B. Sitter, S. Lundgren, et al., Metabolite composition in breast tumors examined by proton nuclear magnetic resonance spectroscopy. Anticancer Res. 19, 1737-1746 (1999). [27] R. M. Sherry, Cancer prevention: role of surgery in cancer prevention, in: V.T. de Vita, S. Hellman, S.A. Rosenberg, Cancer Principles & Practice of Oncology 6th Edition, Lippincott Williams & Wilkins, Philadelphia, 2001, p. 617-626. [28] I.C. Smith, D.E. Blandford, Diagnosis of cancer in humans by 1H NMR of tissue biopsies, Biochem. Cell Biol. 76, 472-476 (1998). [29] J.C. Wallace, G.P. Raaphorst, R.L. Somorjai, C.E. Ng, M. Fung Kee Fung, M. Senterman, I.C. Smith, Classification of 1H MR spectra of biopsies from untreated and recurrent ovarian cancer using linear discriminant analysis, Magn. Reson. Med. 38, 569-576 (1997). [30] E.A. Boss, S.H. Moolenaar, L.F.A.G. Massuger, H. Boonstra, U.F. Engelke, J.G. de Jong, R.A. Wevers, High-resolution proton nuclear magnetic resonance spectroscopy of ovarian cyst fluid, NMR Biomed. 13, 297-305 (2000). [31] L.F.A.G. Massuger, P.B.J. van Vierzen, U. Engelke, A. Heerschap, 1H-magnetic resonance spectroscopy. A new technique to discriminate benign from malignant ovarian tumors, Cancer 82, 1726-1730 (1998). [32] F.V. Coakley, A. Qayyum, J. Kurhanewicz, Magnetic resonance imaging and spectroscopic imaging of prostate cancer, J. Urology, 170, S69-S76 (2003).

317

Chapter 20

Concluding Comments and Outlooks: Prevention, Early Detection & Monitoring Cancer in a more Comprehensive Perspective
_______________________________________________________________________________

In the immediately preceding chapter, we projected some possibilities via optimized MRS and MRSI for improving surveillance of patients with or at risk for specific cancers. The rationale for this approach is to help achieve the positive aims of screening, as outlined in chapter 19. The most important of these, of course, is improved survival, together with less radical therapy when cancer is detected at an earlier stage. It was also noted, however, that with greater diagnostic intensity, there would be a larger number of patients detected with early stage cancer, as well as with precursor lesions. We alluded, as well, to the potentially untoward consequences of such an approach, not the least of which is the impact of receiving a diagnosis of cancer, together with the side effects of therapy, even if less radical than that required for more advanced disease. This raises not only scientific, technological and clinical questions, but also ethical issues concerning the risks and benefits of procedures such as magnetic resonance for cancer screening [1].

20.1 Emphasis upon quality of life

As pointed out by Longo [2], patients experience the diagnosis of cancer as one of the most traumatic events that has ever happened to them. Independent of prognosis, the diagnosis brings with it a change in a persons self image and in his or her role in the home and workplace (p. 491).

318

Increasingly, psychosocial issues are being recognized as an integral part of clinical oncology. In fact, there are entire journals devoted to this area, the Journal of Psychosocial Oncology, e.g., is now in its 21st volume. Quality of life is becoming more and more integrated into all aspects of care of patients with cancer. Rehabilitation in its broadest sense can be considered synonymous with this concern.

20.1.1 The vital importance of return to healthy work With progress in diagnostics and therapy, increasing numbers of patients return to their jobs after, or even during cancer treatment. Return to work is recognized as an important component of quality of life for patients with cancer, helping them maintain not only economic, but often also emotional stability, during treatment and its aftermath [3]. Paraphrasing the late Professor Bertil Gardell [4]: work is one of the most important potential sources of social and psychological well being, which can provide much of the meaning and structure in adult life. Unfortunately, however, for many working people, this potential is far from reality. Instead, the contemporary work environment is all too frequently where employed adults spend the majority of their waking hours performing activities that are characterized as demanding, constraining, and otherwise stressful. Reflecting pressures of global competition, trends in working life are towards increasing job demands, working hours and job instability [5, 6]. Little systematic attention has been given to what kind of work environment the patient with cancer is returning: a healthy job or one which adds yet another major burden to the patients already heavy load of stressors? To approach this question, information about the work environment must first be obtained. Theoretical and methodological advances in occupational psychosocial health could prove invaluable, particularly by identifying potentially modifiable stressors [5, 7]. An objective, comprehensive evaluation of the work environment viewed together with clinical status, can also help avoid the pitfall of oversheltering patients with cancer, by preventing them from performing challenging, but not untowardly taxing tasks. The goal is to guard not only against job overload, but also under-load, which further reinforces the debilitating view that the diagnosis of cancer has irretrievably changed ones role in vital aspects of life such as the workplace. The bio-behavioral model of malignant disease outlines multi-factorial pathways by which psychosocial factors and health behaviours interact

319

with biological processes that, in turn, influence tumor physiology and resistance [8]. Consistent with that model, there are data concerning the deleterious impact of e.g. job strain, poor social well-being, and depression upon helper-inducer T-cells vascular endothelial growth factor, and interleukin-6, inter alia [8, 9]. However, in contrast to strong evidence linking work-related toxic exposures to risk of several cancers [10], and emerging support for an association between night shift work and risk of breast and colon cancer [11, 12], observational studies have yet to provide a coherent picture concerning the etiological relationship between occupational psychosocial exposures and cancer risk, per se, and some null results have been reported [13]. At the same time, substantial empirical data associate untoward working conditions to depression, burnout and other adverse mental health outcomes [5, 14], and stressful work is increasingly accepted as an important risk factor for cardiovascular disease [15, 16]. For patients with cancer, these are co-morbidities of particular relevance. Intervention studies can provide a key link in etiological research. Randomized clinical trials are therefore needed to examine the impact of ameliorating exposure to job stressors upon patients with cancer returning to work, as well as upon markers of importance in cancer development, and upon risk of cancer per se. As a first step, as described in [17], we are working to develop a pro-active approach, whereby clinicians together with patients with cancer formulate tailored, realistic modifications in the work environment. Successful initiatives for improving working life of individual patients can inform wider efforts, in collaboration with other key participants (industrial hygienists, epidemiologists, labour and management), to create a healthy work environment in a broader public health perspective [18]. This approach would offer a new dimension to the concept of occupational rehabilitation of patients with cancer, providing further impetus to its realization. 20.1.2 A patient-centered global and comprehensive strategy We conclude this book by suggesting that there is a need to consider advances in cancer diagnostics as provided by molecular imaging within a broader perspective. Overall, this can be seen as a multilevel patient-centered global and comprehensive strategy in which quality of life, treatment and clinical monitoring are part of an integrated whole.

320

References
[1] S.K. Plevritis, D. M. Ikeda, Ethical issues in contrast-enhanced magnetic resonance imaging screening for breast cancer, Top. Magn. Reson. Imaging 13, 79-84 (2002). [2] D. L. Longo, Approach to the patient with cancer, in: E Braunwald, A. Fauci, D.L. Kasper, S.L. Hauser, D.L. Longo, J.L. Jameson, Harrisons Principles of Internal Medicine, 15th Edition, McGraw-Hill, New York, 2001, p. 491-497. [3] P.N. Schultz, M.L. Beck, C. Stava, R.V. Sellin, Cancer survivors work-related issues, AAOHN 50, 220-226 (2002). [4] B. Gardell, Work organization and human nature, The Swedish Work Environment Fund, Stockholm, 1987. [5] K. Belkic, The Occupational Stress Index: An Approach Derived from Cognitive Ergonomics and Brain Research for Clinical Practice, Cambridge International Science Publishing, Cambridge, England, 2003, ISBN 1-898326-02-9. [6] P. Landsbergis, The changing organization of work and the safety and health of working people: A commentary, J. Occup. Environ. Med. 45, 61-72 (2003). [7] L. Levi, Preventing Work Stress, Reading: Addison-Wesley Co., 1981. [8] S.K. Lutgendorf, E.L. Johnsen, B. Cooper, et al., Vascular endothelial growth factor and social support in patients with ovarian carcinoma, Cancer 95, 808-815 (2002). [9] N. Kawakami, T. Tanigawa, S. Araki et al., Effects of job strain on helper-inducer (CD4+CD29+) and suppressor-inducer (CD4+CD45RA+) T cells in Japanese blue-collar workers, Psychother. Psychosom. 66, 192-198 (1997). [10] H-O. Adami, D. Hunter, D. Trichopoulos, Textbook of Cancer Epidemiology, Oxford University Press, New York, 2002. [11] E.S. Schernhammer, F. Laden, F.E. Speizer, et al., Rotating night shifts and risk of breast cancer in the women participating in the nurses health study, J. Natl. Cancer Inst. 93, 1563-1568 (2001). [12] E.S. Schernhammer, F. Laden, F.E. Speizer, et al., Night-shift work and risk of colorectal cancer in the nurses health study, J. Natl. Cancer Inst. 95, 825-828 (2003). [13] A.J. van Loon, M. Tijhuis, P.G. Surtees, et al., Lifestyle risk factors for cancer: the relationship with psychosocial work environment, Int. J. Epidemiol. 29, 785-792 (2000). [14] L. Levi, More jobs, better jobs and health-policy: challenges to science, Third International Conference on the Work Environment and Cardiovascular Disease, International Commission on Occupational Health, Dsseldorf, 2002. [15] P.L. Schnall, K. Belkic, P.A. Landsbergis, D. Baker, Occupational Medicine: State of the Art Review. The Workplace and Cardiovascular Disease 15 (2000). [16] K. Belkic, P. Landsbergis, P. Schnall, D. Baker, Is job strain a major source of cardiovascular disease risk? Scand. J. Work Environ. Health. 30, 85-128 (2004). [17] K. Belkic, Return to healthy work for patients who have been treated for cancer: a proactive clinical perspective informed by occupational health psychology. In: S. Giga, P. Plaxman, J. Houdmont, M. Ertel, European Academy of Occupational Health Psychology Conference Proceedings Series, ISBN: 0-9539936-3-9, ISSN 1473-0200, I-WHO Publications, Nottingham, England, p. 25, 2003. [18] J. Fisher, K. Belkic, A public health approach in clinical practice, Occupational Medicine: State of the Art Review. The Workplace and Cardiovascular Disease 15, 245-253 (2000).

321

Appendix 1 Glossary of Some Terms Relevant to MRS and MRSI

Absorption Spectrum The positive-definite, real part of the complex spectrum, (under ideal conditions, i.e. no noise, no initial time delay, etc.) Otherwise, it can be constructed from a complex-valued spectrum as e.g. in Ref. [1], see also phasing, in this glossary. Aliasing When the signal frequency is higher than the sampling interval, these higher frequencies e.g. peaks are missed using FFT, and are aliased to lower frequencies, i.e. folded back to the Nyquist interval and disguised as lower frequencies that lead to spectral distortions.

Apodization Multiplication of the acquired FID by a smoothly varying function, e.g. a Gaussian or an exponentially decaying function, in order to suppress the noisier end of the FID. As a consequence of this procedure, the peaks are broadened. Clinical MR scanner (main components) (1) A main magnet, which produces a very strong and homogenous field inside the bore of the magnet, and which must be extremely stable in time. Most clinical scanners operate at 1.5T. Note that a stronger static magnetic field yields improved SNR, but homogeneity is crucial to obtain the benefit of this increased field strength. (2) Gradient coils that are situated inside the bore of the main magnet, and are used to produce magnetic field gradients in the X-, Y- and Z-axes. These can be rapidly varying over time. (3) Radio frequency transmitter/receiver coil mounted within the gradient coil assembly. The received RF signals are then analysed. (4) A couch or gantry in the centre-most part of the scanner, upon which the patient lies.
Dispersion Spectrum Imaginary part of the complex spectrum under the same conditions as for the absorption spectrum, see also comment above in this glossary about the absorption spectrum.

322

Echo Time The time required for a spin-echo sequence, varied by the time delay between the 90 and 180 RF pulses, see spin-echo phenomenon, in this glossary. Short TEs capture metabolites with shorter T2 relaxation times (e.g. myoinositol, glutamine-glutamate, and lipids), whereas at longer TE, many metabolites will have already decayed. Long TEs have been used as a technique for lipid suppression, especially in MRS of the breast, but as a consequence, fewer metabolites are recorded and SNR is poorer. Eddy Currents Current induced in the magnet structure by field gradient pulses, creating additional magnetic fields that add to the static field B0. Zero-order eddy currents will create a frequency-dependent phase shift, whereas first order eddy currents will dephase the spins and thereby decrease SNR. Similarly to magnet inhomogeneities (see shimming), these will distort peak shapes and render spectral quantification more difficult. For further details, including strategies for correcting eddy currents, see Ref. [2]. Gradient-echo pulse sequence The initial RF pulse produces a flip-angle which is < 90, and is rephased by a magnetic gradient to generate an echo signal. Shorter TR and TE can be used and this allows the scan to be obtained much more rapidly [7]. J Coupling The magnetic field of one nucleus influences the external magnetic field sensed by the neighboring nucleus. This is mediated by binding electrons in the bond between the two coupled nuclei (these are weak spin-spin interactions). This results in a spectrum with the resonance of the coupled nucleus split into two lines, such that a doublet of two peaks is seen, e.g. the lactate doublet. The coupling constant denotes the difference in frequency between the two peaks. Magnitude Spectrum Absolute value of the complex spectrum. Nuclear-Overhauser Enhancement Energy transfer between protons and 31P or 13C or other nuclei used to increase the signal intensity. Phasing Whenever the initial phase of an FID is not zero, the real and imaginary parts of the complex spectrum contain mixtures of absorption and dispersion mode spectra. Phasing is the process by which the spectrum is sorted into the real and imaginary spectra, such that: Absorption ( ) = Real( ) cos ( ) + Imaginary ( ) sin ( ) Dispersion ( ) = Imaginary ( ) cos ( ) + Real ( ) sin ( ) For further discussion, see Ref. [2].

323

Power Spectrum The square of the magnitude spectrum. Proton Decoupling Weak interactions between phosphorus atoms and protons in the molecules or in surrounding water lead to a broadening of spectral lines. By proton decoupling, 31P spectrum will contain sharper lines and proportionally higher peak intensities (given that the peak area is constant). Pulse Repetition Time Acquisition time plus recovery interval or sequence delay time. The latter allows spin-lattice relaxation so therefore depends on T1 and the flip angle. Described for NMR with illustration in Ref. [3]. Shimming Correction of magnetic field inhomogeneities, by adjusting static gradients. A key component of preparation for MRS, without which the FID decays rapidly, the signal intensity will appear lower than its true value, and the Lorentzians obtained with transformation into the frequency domain will be widened and distorted. Single-Voxel Techniques Selection of three orthogonal slices such that the region at the intersection of the slices is the source of the signal. See [2, 4-7] for further details. STEAM (STimulated Echo Acquisition Model) = Uses 90 RF pulses to acquire stimulated echo from the confluent voxel, requires gradient suppression, and has lower SNR than PRESS or ISIS (see next two singlevoxel techniques), but requires shorter echo times. Can detect more complicated spin systems such as glutamate and glutamine. PRESS (Point-RESolved Spectroscopy) = Acquires the second echo from a 90- 180 -180RF pulse sequence. PRESS provides improved SNR but previously required longer echo times, although recent advances allow PRESS to be performed at short TE. PRESS is less sensitive to motion than STEAM. ISIS (Image-Selected In Vivo Spectroscopy) = Unlike STEAM and PRESS which can be performed in a single experiment, ISIS requires 8 difference experiments, with an add/subtract scheme to cancel out signals from regions outside the voxel of interest. Useful for 31P MRS since the signals are acquired directly after the detection pulse, see Ref. [5, 7] for further details. Spin-Echo phenomenon Loss of phase coherence (dephasing) means loss of RF signal. Part of this loss is due to spin-spin relaxation, expressed by T2, which is an inherent property of the material. The observed rate of decay of phase coherence T2*, is always faster than T2, because of inhomogeneities of the magnetic field. The part of dephasing due to inhomogeneities can be cancelled by the spin-echo

324

maneuver, in which a 180 RF pulse is applied after termination of the 90 RF pulse. The latter serves to re-gather the transverse relaxation vectors so that they emit a strong signal. For more details and graphic illustrations, see Ref. [8]. Spoiled gradient echo sequence A technique used to dephase the transverse magnetization following signal detection, so that only the longitudinal magnetization contributes to the magnetization vector. This is done to avoid image artefacts and contrast errors in MRI [7, 11]. Susceptibility The degree to which a material becomes temporarily magnetized when placed in a magnetic field. With respect to body tissues, the highest susceptibility is in blood and blood breakdown products since these contain iron. Bone has the lowest susceptibility, and most other tissues are intermediate. At the borders between tissues of different susceptibility, micro-gradients are engendered and dephasing is affected. These create artefacts, especially on gradient-echo and echo-planar imaging [7]. T2 Relaxation Time Reflects the decay of transverse magnetization (Mxy), due to loss of phase coherence among the precessing nuclei (spin-spin relaxation), the time at which transverse magnetization has decayed to 37% of its maximum strength. Two-Dimensional MR Spectroscopy Application of 2D NMR spectroscopic techniques to produce a 2D spectrum, which plots intensity versus intensity on two axes. This can be either chemical shift versus chemic shift (COSY) or chemical shift versus spin-spin coupling frequencies (J resolved spectroscopy) [9].

Zero filling With the Fast Fourier Transform, the acquisition time is often set to record the FID until well past the point where the signal disappears into the noise. Instead of recording a signal with nothing in it but noise, the signal is lengthened by adding zeros. In fact, this does not improve resolution, but merely renders the shape spectrum more visually presentable, see Ref. [10].

References
[1] Dz. Belkic, P.A. Dando, J. Main, H.S. Taylor, Three novel high-resolution nonlinear methods for fast signal processing, J. Chem. Phys. 133, 6542-6556 (2000). [2] D. J. Drost, Proton magnetic resonance spectroscopy in the brain: Report of AAPM MR Task Group #9, Med. Phys. 29, 2177-219 (2002). [3] H. Gunther, NMR Spectroscopy. Basic principles, concepts and applications in chemistry, 2nd Edition, John Wiley & Sons, Chichester, 2001.

325

[4] H. C. Charles, The whole body NMR Spectroscopy examination. Resonance Technologists Educational Seminars 3, 5-18 (2000).

Section for Magnetic

[5] B. Vikhoff-Baaz, In vivo MRS and MRSI. Performance analysis, measurement considerations and evaluation of metabolite concentration images, Gteborg University, Doctoral Dissertation, 2000. [6] E. R. Danielsen, B. Ross, Magnetic Resonance Spectroscopy Diagnosis of Neurological Diseases, Marcel Dekker, Inc., New York, 1999. [7] D.W. McRobbie, E.A. Moore, M.J. Graves, M.R. Prince, MRI from picture to proton, Cambridge University Press, Cambridge, 2003. [8] P. Fleckenstein, J. Tranum-Jensen, Anatomy in Diagnostic Imaging, Blackwell, Copenhagen, 2001. [9] L. Kwock, J.K. Smith, M. Castillo, et al., Clinical applications of proton MR spectroscopy in oncology, Technol. Cancer Res. Treat. 1, 17-28 (2002). [10] Dz. Belkic, Pad-based magnetic resonance spectroscopy, J. Comp. Meth. Sci. Eng. 3, 601760 (2003). [11] M.A. Brown, R.C. Semelka, MRI basic principles and applications. 2nd Edition, John Wiley & Sons, New York, 1999.

327

Appendix 2 List of Acronyms

ADH3 = fast metabolizing allele of alcohol dehydrogenase AIDS = acquired immunodeficiency disorder AMARES = Advanced Method for Accurate Robust and Efficient Spectral fitting ARMA = auto-regressive moving average AT = ataxia-telangiectasia ATP = adenosine triphosphate a.u. = arbitrary units AUB = abnormal uterine bleeding BBB = blood brain barrier BPH = benign prostatic hypertrophy CD = cluster of differentiation CE = contrast enhancement CG = central gland (of the prostate) CHESS = CHEmical Shift Selective Cho = choline CI = confidence intervals CIN = cervical intraepithelial neoplasia CLL = chronic lymphocytic leukemia CNS = central nervous system COSY = 2D correlated spectroscopy Cr = creatine CSI = chemical shift imaging CT = computerized tomography

328

DFI = discretized Fourier integral DFT = discretized Fourier transform DRE = digital rectal exam DRESS = Depth RESolved Spectroscopy DTPA = diethylenetriaminepentaacetic acid DWI = diffusion weighted imaging E = energy EBV = Epstein-Barr virus EC = endometrial carcinoma ECG = electrocardiogram EEG = electroencephalogram EF = enhancement factor FDG = 18F-deoxy glucose FID = free induction decay FFDM = full-field digital mammography FFT = fast Fourier transform FPT = fast Pad transform FSE = fast spin echo FWHM = full-width at half-maximum amplitude GBM = glioblastoma multiforme GI = gastrointestinal Glx = glutamine, glutamate GRE = gradient echo HCC = hepatocellular carcinoma HIV = human immunodeficiency virus H Pylori = Helicobacter Pylori

329

HLSVD = Hankel-Lanczos Singular Value Decomposition HPV = human papilloma virus HRMAS = high-resolution magic-angle spinning HRT = hormone replacement therapy HRV = heart rate variability HSG = hysterosonography HSV = herpes simplex virus HTLV 1 = human T-cell leukemia/lymphoma virus l Hz = hertz ILP = isolated limb perfusion IMRT = intensity modulated radiation therapy ISIS = Image Selective In vivo Spectroscopy Lac = lactate LCModel = Linear Combination of Model in vitro Spectra LHRH = luteinizing hormone releasing hormone Lip = lipids LP = Linear Predictor MALT = mucosa-associated lymphoid tissue MHz = megahertz mI = myoinositol mIG = myoinositol-glycine MRC = MR colonography MRI = magnetic resonance imaging MRS = magnetic resonance spectroscopy MRSI = magnetic resonance spectroscopic imaging MRUI = Magnetic Resonance User Interface NAA = N-acetyl aspartate

330

NHL = non-Hodgkins lymphoma NK = natural killer NMR = nuclear magnetic resonance NTP = nucleotide triphosphate OR = odds ratio PA = Pad approximant PAH = polyaromatic hydrocarbons PC = phosphocholine PCr = phosphocreatine PD = proton density PDE = phosphodiester PE = phosphoethanolamine PET = positron emission tomography PI = pulsatility index Pi = inorganic phosphate PIN = prostatic intraepithelial neoplasia PIP2 = phosphatidyl inositol 4,5-biphosphate PME = phosphomonoester ppm = parts per million PRESS = Point-RESolved Spectroscopy PRIME = Proton Regional Imaging of Metabolites PSA = prostate specific antigen PSAD = prostate specific antigen density PTEN = phosphatase and tensin homolog deleted on chromosome 10 PZ = peripheral zone (of the prostate) QRS = the Q R and S waves which together form the QRS complex, representing depolarization of the ventricles of the heart

331

RF = radiofrequency RI = resistive index rms = root mean square ROPE = Respiratory-Ordered Phase Encoding RSD = renal stone disease RT = radiation therapy Sd = standard deviation Se = standard error of the mean SFM = screen-film mammography SIR = standardized incidence ratio SMM = scintimammography SNR = signal-to-noise ratio SPIO = superparamagnetic iron oxide SSFSE = single shot fast spin echo STEAM = STimulated Echo Acquisition Mode T = units of tesla TDL = tumefactive demyelinating lesions TE = echo time TMA = trimethylamines Tmax = time to reach maximum enhancement TMS = tetramethylsilane TNM = tumor lymph node - metastases TOSCY = 2D total correlation spectroscopy TR = repetition time TRUS = transurethral ultrasound TVS = transvaginal ultrasound (sonography)

332

VARPRO = variable projection method VOI = volume of interest or voxel of interest WHO = World Health Organization