IJSTE - International Journal of Science Technology & Engineering | Volume 3 | Issue 01 | July 2016

ISSN (online): 2349-784X

Improving The Performance of CAD System for

Diagnosis of Diseases from CT Images
T. Primya
Assistant Professor
Department of Computer Science & Engineering
Dr.N.G.P Institute of Technology, Coimbatore.

G. Kanagaraj
Assistant Professor
Department of Computer Science & Engineering
Kumaraguru college of Technology

G. Selva Priya
Assistant Professor
Department of Computer Science & Engineering
Dr.N.G.P Institute of Technology, Coimbatore.

V. Suresh
Assistant Professor
Department of Computer Science & Engineering
Dr.N.G.P Institute of Technology, Coimbatore.

Abstract
Pulmonary embolism (PE) is an extremely common and highly lethal condition that is a leading cause of death in all age groups.
CAD is increasingly used in clinical practice to assist physicians in the detection of subtle abnormalities. Over the past 10 years,
computed tomography (CT) scanners have gained acceptance as a minimally invasive method for diagnosing PE. In this Paper
Computer-aided detection (CAD) scheme is developed for PE detection. It includes Lung segmentation, Candidate Extraction,
Feature Extraction, classification, performance Evaluation. Segmentation is done by means of adaptive threshold and Region
growing process. In Candidate Extraction tobogganing algorithm was applied to detect and extract PE candidates by removing
the thoracic, mediastinum and other nonlung areas. In Feature Extraction the features are extracted by using Grey Level cooccurrence Matrix. Then the extracted features are to be stored in the database. Classification is done by means of k-Nearest
Neighbor classifier which is used to find True Positive and False positive. Then the performance evaluation is measured by using
Receiver Operating Characteristics (ROC) curve.
Keywords: Computer Aided Detection (CAD), Grey Level Coocurrence Matrix (GLCM), k-Nearest Neighbour
_______________________________________________________________________________________________________
I.

INTRODUCTION

Pulmonary embolism (PE) is a sudden blockage in a lung artery. The blockage usually is caused by a blood clot that travels to
the lung from a vein in the leg. A clot that forms in one part of the body and travels in the bloodstream to another part of the
body is called an embolus. PE is a serious condition that can damage part of our lung because of a lack of blood flow to our lung
tissue. This damage may lead to pulmonary hypertension (increased pressure in the pulmonary arteries), Cause low oxygen
levels in your blood, damage other organs in your body because of a lack of oxygen.
If a blood clot is large, or if there are many clots, PE can cause death.PE most often is a complication of a condition called
deep vein thrombosis (DVT). In DVT, blood clots form in the deep veins of the body most often in the legs. These clots can
break free, travel through the bloodstream to the lungs, and block an artery. Deep vein clots are not like clots in veins close to the
skin's surface. Those clots remain in place and do not cause PE.
Blood clots called deep vein thrombi often develop in the deep leg veins. A pulmonary embolism (PE) occurs when clots
break off from the vein walls and travel through the heart to the pulmonary arteries. Although PEs are a highly treatable disease
if detected in a timely manner, they are still the third most common cause of death in the U.S.
Computer-aided detection (CAD) methods could be very useful in supporting radiologists in the identification of early-stage
pathological objects. It has already been demonstrated by several studies that, in addition to a considerable time saving, the
sensitivity of radiologists assisted by computer-aided detection (CAD) systems improves with respect to the performance of the
radiologists alone. A wide range of commercial systems are on the market, each specializing in a particular application area.
Many CAD systems are aimed at the detection of breast, lung, or colon cancer with X-ray, computed tomography (CT) or
magnetic resonance imaging. The design of a CAD system is a difficult and challenging task.
In most applications, the abnormalities to be detected can be in different stages of development or show morphologically
different manifestations of a disease. These factors lead to some CAD systems being able to better detect certain types of
abnormalities. As a result, it is reasonable to expect that it can be beneficial to combine multiple CAD systems for a given task.
In this work we present generic techniques to combine CAD systems and investigate what kind of performance increase can be
expected.

All rights reserved by www.ijste.org

406

Improving The Performance of CAD System for Diagnosis of Diseases from CT Images
(IJSTE/ Volume 3 / Issue 01 / 072)

II. RELATED WORKS

In paper [1] they have proposed two-dimensional LF segmentation algorithm and adapted it to interstitial pneumonia (IP)
patterns. The aim is development of a 2D LF segmentation algorithm adapted to lung affected by interstitial pneumonia (IP)
patterns. 22 CT examinations are used as image dataset. Algorithm used is K-means Clustering Algorithm followed by a filling
operation to obtain an initial LF order estimate. In Lung Field Refinement K-means clustering results in LF under segmentation
due to misassigned IP areas into Surrounding Tissue (ST).It was performed using a SVM classifier based on local texture
analysis. In Texture analysis Gray Level Concurrence Matrix (GLCM) is a well-established tool for characterizing the spatial
distribution of grey levels in the images. In Feature Selection Plus l-take away r selection strategy was implemented. The
method consists of applying sequential forward selection l times followed by r steps of sequential backward selection. Finally
Border pixel labeling based on gray level and wavelet coefficient statistics can accurately segment lung fields in presence of
interstitial pneumonia patterns.
In paper [2] they have proposed a novel lung segmentation technique that can determine the threshold for each CT slice in a
patient stack and automatically do the lung segmentation. The accuracy is 98% when the method was tested on five patient
stacks that contained 914 slices. It is based on Adaptive thresholding for segmenting the lungs from CT images. Advantage is
that segmentation accuracy will be improved by using threshold.
In paper [3] they proposed a new system for the automatic detection of PE in contrast enhanced CT images. The system
consists of candidate detection, feature computation and classification. Candidate detection focuses on the inclusion of PE and
the exclusion of false detections. Feature computation does not only focus on the intensity, shape and size of an embolus, but
also on locations and the shape of the pulmonary vascular tree. Several classifiers have been tested. The system was trained on
38 CT data sets. The sensitivity of the system on the evaluation data is 63% at 4.9 false positives per data set.
III. SYSTEM ARCHITECTURE

Fig. 1: CAD System for detection of PE diseases

Segmentation:
Segmentation is used for partition of lung tissues from CT images. It is done by using Adaptive threshold and region growing
process. The goal of segmentation is to simplify and/or change the representation of an image into something that is more
meaningful and easier to analyze. Image segmentation is typically used to locate objects and boundaries in images. The result of
image segmentation is a set of segments that collectively cover the entire image, or a set of contours extracted from the image.
Adaptive Threshold:
1) Step 1: Select an initial estimate for T
2) Step 2: Segment the image using T. This will produce two groups of pixels. G1 consisting of all pixels with gray level
values >T and G2 consisting of pixels with values <=T.
3) Step 3: Compute the average gray level values mean1 and mean2 for the pixels in regions G1 and G2.
4) Step 4: Compute a new threshold value
T = (1/2)(mean1 +mean2)
5) Step 5: Repeat steps 2 through 4 until difference in T in successive iterations is smaller than a predefined parameter T0.
6) Step 6: complement the threshold image using imcomplement.
7) Step 7: Fill the holes of the images using imfill operation.
8) Step 8: clear the boarder of the image using bwareaopen.
9) Step 9: Fill the pixel values for cleared image from original image.

All rights reserved by www.ijste.org

407

Improving The Performance of CAD System for Diagnosis of Diseases from CT Images
(IJSTE/ Volume 3 / Issue 01 / 072)

Region Growing:
1) Step 1: Read isolated regions of CT images.
2) Step 2: Determine set of pixels(S) inside the region and queue (Q) of pixels
3) Step 3: Define (x0, y0) a pixel inside the region.
Initialize S=
Q= {(x0, y0)}
4) Step 4: Extract pixel P from queue Q.
5) Step 5: Add P to S.
6) Step 6: For each neighbor P' of P:
if P' is "similar" to P and P' S then
add P' to Q.
7) Step 7: If Q= then end, else return to 1.
Candidate Extraction:
After segmenting the images the tobogganing algorithm was applied to detect and extract PE candidates in CT images. The
cluster of every path flowing down to the local minima is the result of tobogganing. It was applied only within the segmented
lung regions having HU values between -70 and 230.
1) Step1: Apply three preprocessing techniques to the image, which are median filter, histogram normalization and histogram
equalization.
2) Step 2: Determine the threshold value.
3) Step 3: Click mouse in the region of interest.
4) Step 4: Choose N N neighbourhood
5) Step 5: Set the initial value for Total of pixels in the region with 1 and Total of grey level for all pixels in the region with
original grey level value of the initial seed pixel.
6) Step 6: Calculate the mean value, x (which is known as region mean) and standard deviation, of the N N neighbourhood
using Equations
=1
=

=1( )2
1
Where i x is grey level of i-th pixel in the N N neighbourhood and n is total of pixels in N N neighbourhood
7) Step 7: Grow the seed pixels to its neighbours pixels. Compare the grey level of the seed pixel with its neighbours pixel.
Include the neighbour pixel into the region if it satisfies one of the conditions listed below.
If the gradient of the pixel is less than 95% of the equalized histogram AND the grey level of the pixel is less or equal to the
preselected threshold.
If the gradient of the pixel is more than or equal to 95% of the equalized histogram AND the grey level of the pixel is not
more than or equal to one standard deviation away from the region mean.
8) Step 8: If the neighbour pixel is included into the region:
Add one (1) to Total of pixels in the region value.
Add original grey level of the neighbour pixel to Total of grey level for all pixels in the region value.
9) Step 9: Set the neighbour pixel, which is added to the region in Step (7) as a new seed pixel.
10) Step 10: Repeat Step (6) to (9) until all pixels have been considered to be grown or the pixel cannot be grown anymore.
11) Step 11: Calculate the value of the size and grey level of the region.
+

Feature Extraction:
Gray Level Co-occurrence Matrix (GLCM) is used to describe the texture features of an image. It considers the spatial
relationship of pixels. Features such as Homogeneity,
Angular Second Moment (ASM), Entropy, Dissimilarity, correlation, contrast, Inverse Difference Moment ( IDM ), Cluster
Shade, Cluster Prominence, Sum of Squares, Variance are to be extracted.
K-NN Classifier:
The k-nearest neighbor algorithm is amongst the simplest of all machine learning algorithms. The k-nearest neighbor algorithm
is a method for classifying objects based on closest training examples in the feature space. k-NN is a type of instance-based
learning, or lazy learning where the function is only approximated locally and all computation is deferred until classification.

All rights reserved by www.ijste.org

408

Improving The Performance of CAD System for Diagnosis of Diseases from CT Images
(IJSTE/ Volume 3 / Issue 01 / 072)

Fig. 2: Flowchart for k-NN

False Positive Reduction:

The main objective of the CAD system and is to obtain a high level of true positives detection rate even for small cells and a low
number of false positives. Based on the feedback of a radiologist in detection the number of True Positives (TP), True Negatives
(TN), False Positives (FP) and False Negatives (FN) are measured.

TN
Specificity

=
TN +FP
TP +TN

Accuracy

=
TP+FP+TN+FN
TP

Precision

=
TP+FP
TP

Recall

=
TP +FN

Where, TP is the identified as True Positive nodules, TN is the True Negative, FP is the number of false positive nodules. FN
is the false negative.
IV. PERFORMANCE EVALUATION
In the training step, optimal PE detection scheme using the training scheme using the training dataset with nine CT examinations
was conducted. Since CAD detects suspicious PE lesions depicted on each individual CT image, one PE lesions can be detected
multiple times by the CAD on different CT slices and have multiple detection scores.
In the testing step, we applied the trained CAD scheme for k-NN classifier. At each evaluation step, the detection scores of TP
and FP lesions cued by CAD scheme were calculated. ROC curve is used for data fitting and analysis program.
V. RESULTS AND DISCUSSIONS
The experiments are conducted on the proposed computer-aided detection systems with the help of real time lung images.

All rights reserved by www.ijste.org

409

Improving The Performance of CAD System for Diagnosis of Diseases from CT Images
(IJSTE/ Volume 3 / Issue 01 / 072)

Fig. 3: original input CT image

Fig. 4: Threshold image

Fig. 5: segmented image

Fig. 6: Extracted Features

All rights reserved by www.ijste.org

410

Improving The Performance of CAD System for Diagnosis of Diseases from CT Images
(IJSTE/ Volume 3 / Issue 01 / 072)

Fig. 7: classification result

Fig. 8: ROC curve

VI. CONCLUSION AND FUTURE WORK

We developed and tested a new CAD scheme for detecting PEs depicted on CT images. To improve CAD performance including
sensitivity and specificity detection we investigated more approaches. In future work many CAD systems are to be developed
and it is to be combined using combination subsystem. Combination results in a large and significant increase in performance
when compared to the best individual CAD system.
REFERENCES
[1]
[2]

Panayiotis Korfiatis, Christina Kalogeropoulou , Anna Karahaliou, Texture classification-based segmentation of lung affected by interstitial pneumonia in
high-resolution CT,Medical Physics,Vol.35,no.12,pp.5290-5302,2008.
Sang Cheol Park, Jun Tan, Xingwei Wang, Dror Lederman Joseph K Leader, Soo Hyung Kim and Bin Zheng, Computer-aided detection of early
interstitial lung diseases using low-dose CT images,Vol.56,no.11,pp.1139-1153,2011.

All rights reserved by www.ijste.org

411

Improving The Performance of CAD System for Diagnosis of Diseases from CT Images
(IJSTE/ Volume 3 / Issue 01 / 072)
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]

Giuseppe Coppini, , Stefano Diciotti, , Massimo Falchini, Natale Villari, and Guido Valli, Neural Networks for Computer-Aided Diagnosis: Detection of
Lung Nodules in Chest Radiograms, IEEE transactions on information technology in biomedicine, vol. 7, no. 4,pp.344-357, Dec 2003.
C. Zhou, H. P. Chan, S. Patel, P. N. Cascade, B. Sahiner, L. M. Hadjiiski,and E. A. Kazerooni, Preliminary investigation of computer-aided detection of
pulmonary embolism in 3D computed tomographic pulmonary angiography (CTPA) images, Acad. Radiol., vol. 12, pp. 782792, 2005.
Sang Cheol Park, Won Pil Kim , Bin Zheng , Joseph K. Leader , Jiantao Pu1 , Jun Tan , and David Gur, Pulmonary Airways Tree Segmentation from CT
Examinations Using Adaptive Volume of Interest,Medical Imaging.Radiol,Vol.7259,pp. 1256-1276,2009.
Georgia D. Tourassi , Erik D. Frederick , Carey E. Floyd Jr, R. Edward Coleman, Multifractal texture analysis of perfusion lung scans as a potential
diagnostic tool for acute pulmonary embolism, pergamon. Department of Radiology, Vol.31, pp.15-25, 2000.
Henri Bouma, Jeroen J. Sonnemans, Anna Vilanova, and Frans A. Gerritsen, Automatic Detection of Pulmonary Embolism in CTA Images, IEEE
transactions on medical imaging, vol. 28, NO. 8, pp.1223-1230, 2009.
Lilla Boroczky, Luyin Zhao, and K. P. Lee, Feature Subset Selection for Improving the Performance of False Positive Reduction in Lung Nodule CAD,
IEEE transactions on information technology in biomedicine, vol. 10, NO. 3, pp.504-511, 2006.
M. Gomathi and Dr. P. Thangaraj, Lung Nodule Detection using a Neural Classifier, IACSIT International Journal of Engineering and Technology,
Vol.2, No.3, pp.291-295, 2010.
Sang Cheol Park, Brian E. Chapman, and Bin Zheng, A Multistage Approach to Improve Performance of Computer-Aided Detection of Pulmonary
Embolisms Depicted on CT Images: Preliminary Investigation, IEEE transactions on biomedical engineering, vol. 58, no. 6, pp.1519-1527, 2011.
S. C. Park, J. Pu, and B. Zheng, Improving performance of computeraided detection scheme by combining results from two machine learning classifiers,
Acad. Radio., vol. 16, pp. 266274, 2009.
Z. V. Maizlin, P. M. Vos, M. C. Godoy, and P. L. Cooperberg, Computer aided detection of pulmonary embolism on CT angiography: Initial experience,
J. Thorac. Image, vol. 22, pp. 324329, 2007.
Engelke, S.Schmidt, A. Bakai, F.Auer,and K.Marten,Computer-assisted detection of pulmonary embolism: Performance evaluation in consensus with
experienced and inexperienced chest radiologists, Eur.Radiol., vol. 18, pp. 298307, 2008.
Lilla Boroczky, Feature Subset Selection for Improving the Performance of False Positive Reduction in Lung Nodule CAD, IEEE transactions on
information technology in biomedicine, vol. 10, no. 3,pp.504-511, July 2006.
Antony Brownea , Brian D. Hudson , David C. Whitleyb , Martyn G. Fordb,Philip Pictonc, Biological data mining with neural networks: implementation
and application of a flexible decision tree extraction algorithm to genomic problem domains, Elsevier, 2003.
M. Gomathi, P. Thangaraj, A Computer Aided Diagnosis System for Lung Cancer Detection using Machine Learning Technique, European Journal of
Scientific Research, 2011.

All rights reserved by www.ijste.org

412