Marketing

An Efficient Brain Tumor Extraction from MRI Images using Entropy Measures
2014-15
Chapter1: Introduction
Image processing is an active research area in which medical image processing is a
highly challenging field techniques are used to process of creation of visual
representation of the inner portions of the human body for medical diagnosis.
Information is conveyed through images. Image processing is a process where input
image is processed to get output also as an image. Main aim of all image processing
techniques is to recognize the image or object under consideration easier visually. All
the images used in todays world are in the digital format. Medical images are images
that show the physical attributes distribution. Medical imaging modalities as in MRI,
CT scan mostly depend on computer technology to generate or display digital images of
the internal organs of the human body which helps the doctors to visualize the inner
portions of the body. CT scanner, Ultrasound and Magnetic Resonance Imaging took
over conventional x-ray imaging, by allowing the doctors see the body's third
dimension. Brain tumor is a serious life altering disease condition. Image segmentation
plays a significant role in image processing as it helps in the extraction of suspicious
regions from the medical images. For this the segmentation of brain MRI images are
done for detection of Tumor location. Magnetic Resonance Imaging is an important
diagnostic imaging technique utilized for early detection of abnormal changes in tissues
and organs. Tumor is an uncontrolled growth of tissues in any part of the body. Most
Research in developed countries show that the number of people who have brain tumors
were died due to the fact of inaccurate detection. However this method of detection
resists the accurate determination of stage &size of tumor. To avoid that a computer
aided method for segmentation (detection) of brain tumor based on the various entropy
measures has been used.
1.1 Brain anatomy overview
BRAIN, the central processing unit of human body, is a soft, delicate and spongy mass
of tissues. It is a steady place for signals to enter and being processed. The brain directs
the things we choose to do (like walking and talking) and the things our body does without
thinking (like breathing). The brain is also in charge of our senses (sight, hearing, touch, taste,
and smell), memory, emotions, and personality. The human brain which functions as the
center for the control of all the parts of human body is a highly specialized organ that
allows a human being to adapt and endure varying environmental conditions. The
Poornima University, Jaipur
M. Tech. (CE)
Page 1
2014-15
human brain enables a human to articulate words, execute actions, and share thoughts
and feelings. In this section the tissue structure and anatomical parts of the brain are
described to understand the purpose of this study.
So, before dealing with the brain tumors and MRI Images, the basic structure of human
brain must be well understood.
Fig. 1.1 The Major Subdivision Of Human Brain [19]
Following are the main parts of the human brain:

1.1.1 Brainstem: It is the lower extension of the brain where it connects to the spinal
cord. Neurological functions located in the brainstem include these necessary for
survival (breathing, digestion, heart rate, blood pressure) and for arousal (being awake
and alert). The brainstem is the pathway for all fiber tracts passing up and down from
peripheral, nerves and spinal cord to the highest parts of the brain.
1.1.2 Cerebellum: The portion of the brain (located at the back) which helps coordinate
movement (balance and music coordination).
1.1.3 Frontal Lobe: It is the front part of the brain and is involved in planning,
organizing, problem solving, selective attention, personality and a variety of higher
cognitive functions including behavior and emotions.
1.1.4 Occipital Lobe: It is the region in the back of the brain which processes visual
information. It also contains association areas that help in the visual recognition of
shapes and colors.
M. Tech. (CE)
Page 2
2014-15
1.1.5 Parietal Lobe: One of the two parietal lobes of the brain are located behind the
frontal lobe at the top of the brain.
1.1.6 Temporal Lobe: There are two temporal lobes, one on each side of the brain
located at above the level of ears. These lobes allow a person to tell one smell from
another. They also help in sorting new information and are believed to be responsible
for short term memory.
The human brain which functions as the center for the control of all the parts of human
body is a highly specialized organ that allows a human being to adapt and endure
varying environmental conditions. The human brain enables a human to articulate
words, execute actions, and share thoughts and feelings.
1.2 Brain Tumors

Under certain conditions, brain cells grow and multiply uncontrollably because for some
reasons the mechanism that control normal cells is unable to regulate the growth of the
brain cells. The abnormal mass of brain tissue is the brain tumor that occupies space in
the skull and interrupts the normal functions of brain and creates an increasing pressure
in the brain. Due to increased pressure on the brain, some brain tissues are shifted,
pushed against the skull or are responsible for the damage of the nerves of the other
healthy brain tissues.
Scientists have classified brain tumor according to the location of the tumor, type of
tissue involved whether they are noncancerous or cancerous. The site of the origin
(primary of secondary) and other factors involved. World Health Organization (WHO)
classified brain tumor into 120 types. This classification is done on the basis of the cell
origin and the behavior of the cell from less aggressive to more aggressive behavior.
Even, some tumor types are graded ranging from grade I (less malignant) to grade IV
(more malignant). This signifies the rate of the growth despite of variations in grading
systems which depends on the type of the tumor.
Grade I: The tissue is benign. The cells look nearly like normal brain cells, and they
grow slowly. This type of brain Tumors are rare in adults.
Grade II: The tissue is relatively slow growing and sometimes spread to nearby normal
tissues and become malignant. The cells look less like normal cells than do the cells in a
Grade I tumor.
Grade III: These are the malignant tissue and have cells that look very different from
normal cells. The abnormal cells are actively growing (anaplastic).
M. Tech. (CE)
Page 3
2014-15
Grade IV: These cover the most malignant tissue and have cells that look most
abnormal and tend to grow quickly. Tumor forms new blood vessels to maintain rapid
growth.
Primary brain tumors are the tumors that originated in the brain and are named for the
cell types from which they originated. They can be benign (non-cancerous) and
malignant (cancerous). Benign tumors grow slowly and do not spread elsewhere or
invade the surrounding tissues. However, occupying a short space, even the less
aggressive tumor can exercise much pressure on the brain and makes it dysfunctional.
Conversely, more aggressive tumors can grow more quickly and spread to other tissues.
Each of these tumors has unique clinical, radiographic and biological characteristics.
Secondary brain tumors originate from another part of the body. These tumors consist of
cancer cells somewhere else in the body that have metastasized or spread to the brain
The most common cause of secondary brain tumors are: lung cancer, breast cancer,
melanoma, kidney cancer, bladder cancer, certain sarcomas, and testicular and germ cell
tumors.
1.3 MRI Images

Raymond V. Damadian invented MRI in 1969 and was the first person to use MRI to
investigate the human body. Eventually, MRI became the most preferred imaging
technique in radiology because MRI enabled internal structures be visualized in some
detail. In MRI signal processing considers signal emissions. The imaging process does
not involve the use of ionizing radiation and hence does not have the associated
potential harmful effects. It is a tomographic imaging technique that produces images of
internal physical and chemical characteristics of an object from externally measured
Nuclear Magnetic Resonance (NMR) signals. MR imaging is based on the well-known
NMR phenomenon. MR signals used for image formation come directly from the object
itself. With MRI, good contrast between different soft tissues of the body can be
observed. This makes MRI suitable for providing better quality images for the brain, the
muscles, the heart and cancerous tissues compared with other medical imaging
techniques, such as Computed Tomography (CT) or X-rays. Magnetic Resonance
Imaging (MRI) used in radiology to investigate the anatomy and physiology of the body
in both health and disease. MRI scanners use magnetic fields and radio waves to form
images of the body. The technique is widely used in hospitals for medical diagnosis,
staging of disease or for accurate detection, shape and size of the tumor.
M. Tech. (CE)
Page 4
2014-15
Figure1.3 shows that a patients head is examined from three plans in a clinical
diagnosis which are axial plane, sagittal plane and coronal plane.
Fig. 1.3Patients Head Examined in a Clinical Diagnosis [21]
The above figure 1.3 shows that the patients head is examined from three planes in a
clinical diagnosis which are Axial plane, Sagittal Plane, Coronal Plane. Furthermore,
T1-weighted brain MRI Images from various planes are shown below.
Fig. 1.3.1 (a) Axial Plane
(b) Sagittal Plane
(c) Coronal Plane [21]
The above figure 1.3.1 shows that the patients head is examined from three planes in a
clinical diagnosis which are Axial plane, Sagittal Plane, Coronal Plane.
Generally, T scan or MRI that is directed into intracranial cavity produces a complete
image of brain. This image is visually examined by the physician for detection and
diagnosis of brain tumor. However this method of detection resists the accurate
determination of stage & size of tumor. To avoid that, this project uses computer aided
method for segmentation (detection) of brain tumor based on the combination of two
algorithms. This method allows the segmentation of tumor tissue with accuracy and
M. Tech. (CE)
Page 5
2014-15
reproducibility comparable to manual segmentation. In addition, it also reduces the time

for analysis. At the end of the process the tumor is extracted from the MRI Image and
its exact position and the shape also determined. The stage of the tumor is displayed
based on the amount of area calculated from the cluster.
1.4 The Image Segmentation

In computer vision, image segmentation is the process of partitioning a digital image
into multiple segments (sets of pixels, also known as super pixels). 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. Segmentation is typically used
to locate objects and boundaries (lines, curves, etc.) in images. More precisely, image
segmentation is the process of assigning a label to every pixel in an image such that
pixels with the same label share certain characteristics. 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 (see edge detection). Each of the pixels in a region is
similar with respect to some characteristic or computed property, such as color,
intensity, or texture. Adjacent regions are significantly different with respect to the same
characteristic. When applied to a stack of images, typical in medical imaging, the
resulting contours after image segmentation can be used to create 3D reconstructions
with the help of interpolation algorithms. This is typically used to identify objects or
other relevant information in digital images.
The purpose of segmentation is to separate image information into clear meaningful
parts by placing boundaries separating the area of the healthy brain from the area of the
cancerous and tumorous brain. The objective of image segmentation is to cluster pixels
into prominent image region. Segmentation must not allow regions of the image to
overlap. Image segmentation can be classified into three categories: spatial clustering,
split and merge schemes, and region growing schemes.
1.5 Application of Image Segmentation

There are various applications of the image segmentation that are described below:
1.5.1 Content Based Image Retrieval: Content-Based Image Retrieval (CBIR), also
known as Query. By Image Content (QBIC) and Content-Based Visual Information
Retrieval (CBVIR) is the application of computer vision techniques to the image
retrieval problem, that is, the problem of searching for digital images in large databases.
M. Tech. (CE)
Page 6
2014-15
Content-based means that the search analyzes the contents of the image rather than
the metadata such as keywords, tags, or descriptions associated with the image. The
term content in this context might refer to colors, shapes, textures, or any other
information that can be derived from the image itself. CBIR is desirable because
searches that rely purely on metadata are dependent on annotation quality and
completeness.
1.5.2 Machine Vision: Machine Vision (MV) is the technology and methods used to
provide imaging-based automatic inspection and analysis for such applications as
automatic inspection, process control, and robot guidance in industry.
1.5.3 Medical Imaging: Medical imaging is the technique and process of creating
visual representations of the interior of a body for clinical analysis and medical
intervention. Medical imaging seeks to reveal internal structures hidden by the skin and
bones, as well as to diagnose and treat disease. Medical imaging also establishes a
database of normal anatomy and physiology to make it possible to identify
abnormalities. Although imaging of removed organs and tissues can be performed for
medical reasons, such procedures are usually considered part of pathology instead of
medical imaging. Locate tumors and other pathologies.
Measure tissue volumes
Diagnosis, study of anatomical structure
Surgery planning
Virtual surgery simulation
Intra-surgery navigation
1.5.4 Object Detection: Object detection is a computer technology related to computer

vision and image processing that deals with detecting instances of semantic objects of a
certain class (such as humans, buildings, or cars) in digital images and videos. Wellresearched domains of object detection include face detection and pedestrian detection.
Object detection has applications in many areas of computer vision, including image
retrieval and video surveillance.
Pedestrian detection
Face detection
Brake light detection
Locate objects in satellite images (roads, forest, crops, etc.)
M. Tech. (CE)
Page 7
2014-15
1.6 Introduction to Brain Tumor Segmentation

Brain Tumor Segmentation is the process of partitioning a digital image into multiple
segments (sets of pixels, also known as super pixels). 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 (lines, curves, etc.) in images. More precisely, image segmentation is
the process of assigning a label to every pixel in an image such that pixels with the same
label share certain characteristics. Segmentation may also depend on various features
that are contained in the image. It may be either color or texture. Before de-noising an
image, it is segmented to recover the original image. The main motto of segmentation is
to reduce the information for easy analysis. Segmentation is also useful in Image
Analysis and Image Compression. In order to investigate and present the image
segmentation, different entropy measures for threshold selection purpose is used.
1.7 Difficulties in Segmentation of Brain MRI

Cortical segmentation has not been made fully automated and operated at high speed
because of the reliability of the MRI with regards to the homogeneity of its magnetic
field. The problems of MRI include:
1.
Noise: random noise connected with MR imaging system. This is known to have
a Rician distribution.
2.
Intensity inhomogeneity also called bias field or shading artifact: the nonuniformity in the radio frequency (RF) field during data collection results in
shading effect.
3.
Partial volume effect: In this type of effect more than one type of class or tissue
occupies one pixel or voxel of an image. The pixels or voxels are called mixels.
Segmentation of MRI outputs are normally done by medical experts and requires
processes which consume time. As the images of tumor tissues from different patients
contain many diverse appearance and gray level intensities and frequently look similar
to normal tissues, the process of automation for segmentation of MRI outputs faces
many challenges. One of these challenges is overcome by utilizing prior information
about the appearance of normal brain when performing classification from a multidimensional volumetric features set. This is tantamount to using a statistical model for
tumor and normal tissue of the same feature set.
M. Tech. (CE)
Page 8
2014-15
1.8 Thresholding
Thresholding is the simplest method of image segmentation. Thresholding is a process
of cinverting a grayscale input image to a bi-level image by using an optimal threshold.
The simplest thresholding methods replace the each pixel in an image with a black pixel
if the image intensity Ii,j is less than some fixed constant T that is, Ii,j<T or a white pixel
if the image intensity is greater than that constant.
1.9 Purpose of Thresholding

The purpose of thresholding is to extract those pixels from some image which represent
an object (either text or other line image data such as graphs, maps). Though the
information is binary the pixels represent a range of intensities. Thus the objective of
binarization is to mark pixels that belong to true foreground regions with a single
intensity and background regions with different intensities.
1.10 Thresholding Algorithm

There are basically two types of the thresholding algorithm the global thresholding
algorithm and the local thresholding algorithm.
1.10.1 Global thresholding algorithms: In global thresholding, a single threshold for
all the image pixels is used. When the pixel values of the components and that of
background are fairly consistent in their respective values over the entire image, global
thresholding could be used.
1.10.2 Local or adaptive thresholding algorithms: In adaptive thresholding, different
threshold values for different local areas are used.
1.11 Thresholding Methods

Other than the thresholding algorithm there are different thresholding methods that are
described below.
1.11.1 Histogram Shape-Based Methods: In this peaks, valleys and curvatures of the
smoothed histogram are analyzed.
1.11.2 Clustering Based Methods: The gray-level samples are clustered in two parts as
background and foreground object.
M. Tech. (CE)
Page 9
2014-15
1.11.3 Entropy Based Methods: It gives the result in algorithms that use the entropy of
the foreground and background regions and the cross-entropy between the original and
binarized image, etc.
1.11.4 Object Attribute-Based Methods: It searches a measure of similarity between
the gray-level and the binarized images, such as fuzzy shape similarity, edge
coincidence, etc.
1.11.5 Spatial Methods: It uses higher-order probability distribution and/or correlation
between pixels.
1.11.6 Local Methods: It adapts the threshold value on each pixel to the local image
characteristics. In these methods, a different T is selected for each pixel in the image.
Now-e-days, speed of computation is no longer an issue for researchers. Therefore, the
focus is directed toward improvement of information from images obtained through the
slice orientation and perfecting the process of segmentation to get an accurate picture of
the brain tumor.
Chapter 2 deals with details of information extracted from research paper related to brain
tumor detection, segmentation, enhancement and feature extraction and its analysis
leading to strengths, weaknesses and gaps in the published research.
Chapter 3 discusses the theoretical aspects of the targeted work and design details of the
technique used by the researchers and also the main approach used in this dissertation.
Chapter 4 elaborates on the proposed system design and its implementation. It includes
details of different tools and techniques and the simulation of the software used. The
architecture description is presented to support the process flow of the work in thesis.
Chapter 5 deals with the experimental results and analysis of the brain tumors through
MRI image segmentation. It also includes the comparison between Shannon and NonShannon entropy measure along with the current comparison with status of the
researcher performance.
Finally, the conclusion is made in chapter 6, summarizing the challenges, proposed
technique and its effectiveness as compared to earlier work.
Next chapter shall discuss about the comprehensive review of the methods and
techniques used to detect brain tumor through MRI image segmentation leading to
problem statement and objective.
M. Tech. (CE)
Page 10
2014-15
Chapter2: Literature Review

A literature review is necessary to identify a problem and the stages of its solution. A
proper literature review provides a solid background for a noble research work.
Literature survey includes the study of various sources of literature in the area of
research. It includes finding the related material from magazines, books, research
articles, scientific research papers published in various conferences, journals &
transactions. Study and understanding the literature other than scientific research papers
is a bit easy as it elaborates the concepts in simple and explanatory ways. At the same
time these contents cannot be considered as base to arrive at the conclusion of framing
research objectives as it is not supported through proper review by various researchers
working in the area. Review of a scientific research paper is a tedious work. It needs the
prior knowledge of the area of research. The scientific research papers are highly
structured, compact and precise in explanation. The researchers need to adopt a certain
path for doing literature review of such literature. There has been many procedures and
processes defined by the researchers to undergo through and arrive at certain conclusions
of research objectives. The five stages of the review process adopted are discussed in
this chapter. In the literature survey, 31 papers of journals and conferences ranging from
the year 1998 to 2015 and have been reviewed. The categorical review related to the
image segmentation and brain tumor detection has been done. Further including the
gaps in the published research, issue wise solution approaches with common finding,
strengths, weaknesses problem statement and objectives are also presented.
2.1 Review Process Adopted

The process diagram is shown in Fig.2.1, which includes in all five stages. The review
process is divided into five stages in order to make the process simple and adaptable by
every researcher. As it reflects from the literature that while beginning the finding of
research objectives, it is necessary to start with a broader domain of any area and
subarea of interest and narrow down to the specific issue, the process described in the
diagram includes the narrowing down along with the research objectives as outcome
with justification of the problem.
2.1.1 Stages of Review Process
Stage 0: Get the Feel.
Stage 1: Get Big Picture.
M. Tech. (CE)
Page 11
2014-15
Stage 2: Get the Details.

Stage 3: Evaluate the Details.
Stage 3+:Synthesize the Details.
Fig. 2.1 Review Process Stage Analysis
The details of various stages followed are discussed below.

2.1.1 Stage 0: Get a feel
This stage is the beginning of the literature review process where in one has to broadly
select the area of interest and start searching the scientific research papers from valid
sources. More than 70 papers have been passed through this process. The whole
dissertation concept is related only on the image segmentation, image enhancement and
brain tumor detection and the algorithms that were developed on brain tumor
segmentation.
2.1.3 Stage 1:Get the big picture
In order to understand the paper broadly and get an idea whether the paper exactly
belongs to the research area or subareas elected or it deviates, if deviates how much,
these concepts are made clear this stage, known as Get Big Picture. In this stage the
author knowing about the problem which was the author's attempt to solve. The solution
value and the knowing result are telling about the criteria of the technique was having. In
that step the various types of brain tumor detection techniques, image enhancement and
image segmentation are discussed. After finding the all solution approaches it was found
that the speed of computation is no longer an issue for researchers and the focus is
directed towards improvement of information from image obtained through the accurate
picture of brain tumor. Total 33 research papers could be selected through applying this
M. Tech. (CE)
Page 12
2014-15
process on more than 70 papers.

2.1.3Stage2: Get the details
Stage2 deals with going in depth of each research paper and understand the details of the
methodology used, problem justification to significance & novelty of the
problem/solution approach exist, precise question addressed, key contribution, scope &
limitations of the work presented.
After reading all the paper related to the tumor detection, image segmentation it was
found that the many papers are totally focused on the noise reduction technique, some
papers are related to segmentation to get an accurate picture, automation of brain tumor,
computation process, decision making technique. By using accurate diagnosis of brain
tumor, proper brain segmentation is required to be used to carry out an improved
diagnosis and treatment.
2.1.4 Stage 3:Evaluate the details
This stage provides an insight how to deal with evaluation of the details presented by
the researchers. In the beginning, it is not possible to evaluate these details, as it needs
to check in relative and comparative aspects in the domain of topic. This stage
evaluates the details in relation to the significance of the problem, novelty of the
problem, significance of the solution, novelty in approach, validity of claims, etc.
Here the detail about the detecting brain tumor through Magnetic Resonance Images
has been explored and comparative study has been carried out.
2.1.5
Stage 3+:Synthesize the detail
Stage3+ deals deals with synthesis of the data, concept & the results presented by the
authors. This stage does not only require the understanding of all research work related
to the topic, but also requires creative thinking and good knowledge in the subarea of
research. Here all possible situations in general needed to be considered while
generalizing.
The brain tumor detection required image segmentation through MRI, to produce
images of soft tissues of human body. For this purpose, the brain is partitioned into two
distinct regions. This is considered to be the most important but difficult part of the
process of detecting brain tumors. A variety of algorithms were developed for
segmentation of MRI images by using different tools and techniques. However here
M. Tech. (CE)
Page 13
2014-15
presents the detection through the entropy measures and then the analysis on the basis
that which will provide the best result.
2.2 Categorical Reviews in An efficient Brain Tumor extraction from

MRI Images using Entropy Measures
After an exhaustive literature review, there have been different types of approaches for
solving the power problems in the circuit. Total 53 papers were reviewed belonging to
conference, journal and transaction publications ranging from year 2001 to 2014. All the
papers were related to single issue of Brain Tumor Detection.
Table 2.2 Categorical Review of Research Papers
S.NO. Name of Issue
Total Paper Number of Paper Reviewed

Reviewed
Conf.
Journal
Transaction
of 13
13
Tumor 32
26
Image restoration
Enhancement
image
Brain
detection from MRI

image
2.2.1 Review outcome in the issue An efficient Brain Tumor extraction from MRI
Images using Entropy Measures:
Most of the papers were centered on a particular issue that was preserving the original
colors indifferent segments of the original image. The approach for color image
segmentation, which approximately preserves the colors in different segments, was
presented through various mechanisms for different entropy measures. The summaries
of the review are shown below.
[M. Stella Atkins, et-al, 1998] has developed a robust fully automatic method for
segmenting the brain from head magnetic resonance (MR) images, which works even in
the presence of radio frequency (RF) inhomogeneities. This method used an integrated
approach which employs image processing techniques based on anisotropic filters and
contouring techniques. The algorithm consists of three incremental steps. The first step
used histogram analysis to localize the head, providing a region which must completely
surround the brain. The second step used nonlinear anisotropic diffusion and automatic
thresholding to create a mask that isolates the brain within the detected head region.
M. Tech. (CE)
Page 14
2014-15
Using the mask as a seed, the final step employs an active contour model algorithm to
detect the intracranial boundary. It was the multistage process, involving first removal
of the background noise leaving a head mask, then finding a rough outline of the brain,
then refinement of the rough brain outline to a final mask [4].
[C.C. Leung, et-al, 2003] has proposed a new approach to detect the boundary of brain
tumor based on the generalized fuzzy operator (GFO). The GFO map all the pixels in
the original image and all the values were passed to the new fuzzy set. The nonhomogeneities density tissue of the brain with tumor could result in achieving the
inaccurate location in any boundary detection algorithms. In MRI images the Boundary
detection with brain tumor was an important image processing technique. It was a
simple but effective method using in boundary detection. The special properties of this
GFO searched the boundary in high accuracy and obtained the fine edge based on its
generalized fuzzy set. The fuzzy sets separate the edge, normal tissue and tumor
respectively [16].
[M. Sasikalal, et-al, 2005] has proposed a computer based system for the detection of
glioblastoma multiform in brain images and compared feature selection algorithms for
the detection of glioblastoma multiform in brain images. Texture features were
extracted from normal and tumor regions (ROI) using spatial gray level dependence
method and wavelet transform. A very difficult problem in classification techniques was
the choice of features to distinguish between classes. The feature optimization problem
was addressed using a Genetic Algorithm (GA) as a search method. Principal
component analysis, classical sequential methods and floating search algorithm were
compared against the genetic approach in terms of the best recognition rate achieved.
The normal and tumor images were classified with an accuracy of97.3% using the entire
feature set [22].
[Phooi Yee Lau, et-al, 2005] has proposed an analytical method to detect tumors in
digitized medical images for 3D visualization. The EGH parameters were used in a
supervised block of input images. These feature blocks were compared with
standardized parameters (derived from normal template block) to detect abnormal
occurrences, e.g. image block which contain lesions or tumor cells. The abnormal
blocks were transformed into three-dimension space for visualization and studies of
robustness. Experiments were performed on different brain disease based on single and
M. Tech. (CE)
Page 15
2014-15
multiple slices of the MRI dataset. The method was effectively capable of identifying
tumor areas in T2-weighted medical brain images taken under different clinical
circumstances and technical conditions, which were able to show a high deviation that
clearly indicated abnormalities in areas with brain disease [19].
[Satish Chandra, et-al, 2009] has proposed a modified version of PSO based algorithm
for the classification of MRI images. The algorithm finds the centroids of number of
clusters, where each cluster groups together brain tumor patterns, obtained from MRI
Images. The results obtained for three performance measures were compared with those
obtained from Support Vector Machine (SVM). The performance analysis shows that
qualitative results obtained from the proposed model were comparable with those
obtained by SVM.
However, in order to obtain better results from the proposed
algorithm it was needed to carefully select the different values of PSO control
parameters [5].
[Tao Wang, et-al, 2009] has proposed a new approach called the fluid vector flow
(FVF) active contour model to address problems of insufficient capture range and poor
convergence for concavities. With the ability to capture a large range and extract
concave shapes, FVF demonstrates improvements over techniques like gradient vector
flow, boundary vector flow, and magnetostatic active contour on three sets of
experiments: synthetic images, pediatric head MRI Images, and brain tumor MRI
images from the internet brain segmentation repository. Experiment on synthetic images
and head MRI images shown that FVF approach had produced better results.
Quantitative experiments on brain tumor images presented that FVF has the largest
mean (0.61) and median (0.60) with smallest standard deviation (0.05) using TM.
Mixed effects model with random data and test effects was used to statistically compare
the differences between FVF and other methods [27].
[M. Murugesan, et-al, 2009] has proposed an automated system for efficient detection
of brain tumor in EEG signals using Artificial Neural Networks (ANNs). The proposed
system had taken an EEG signal with artifacts as input. Firstly, the inputted EEG signal
was subjected to artifacts removal by means of wiener filtering. Then, features of
interest for brain tumor detection were extracted from the EEG signal. EEG signal were
extracted using spectral estimation. Specifically, spectral analysis was achieved by
using Fast Fourier Transform that extracted the signal features buried in a wide band of
M. Tech. (CE)
Page 16
2014-15
noise. The clean EEG data thus obtained was used as training input to the feed forward
back propagation neural network. The trained feed forward back propagation neural
network when fed with a test EEG signal, effectively detected the presence of brain
tumor in the EEG signal. The experimental results demonstrated the effectiveness of the
proposed system in artifacts removal and brain tumor detection [17].
[Ahmed Kharrat, et-al, 2009] has proposed an efficient detection of brain tumor based
on mathematical morphology, wavelet transform and K-means technique. The method is
composed of three steps: enhancement, segmentation and classification. The algorithm
reduces the extraction steps through enhancing the contrast in tumor image by
processing the mathematical morphology. The segmentation and the localization of
suspicious regions were performed by applying the wavelet transforms. To improve the
quality of images and limit the risk of distinct regions fusion in the segmentation phase
an enhancement process was applied. A mathematical morphology was adopted to
increase the contrast in MRI images. Then Wavelet Transform in the segmentation
process was applied to decompose MRI images. At last, the k-means algorithm was
implemented to extract the suspicious regions or tumors [13].
[T.Logeswari, et-al, 2010] has proposed a segmentation method for detection of brain
tumor consisting of two phases. In the first phase, the MRI brain image was acquired
from patients database, In that film artifact and noise were removed. After that
Hierarchical Self Organizing Map (HSOM) was applied for image segmentation. The
HSOM was the extension of the conventional self-organizing map used to classify the
image row by row. In this lowest level of weight vector, a higher value of tumor pixels,
computation speed was achieved by the HSOM with vector quantization [15].
[Somojit Saha, et-al, 2010] has proposed an automated segmentation algorithm of the
head contour and the entire brain. The approach used a high resolution MR image
acquisition protocol for better visualization of normal gray structures of brain. Using
signal nulling effect the author has emphasized enhancing the intensity difference
between white and gray matters in the reconstructed MR image. This leads to the
generation of a histogram where pixel values of different anatomical structures were
distributed around separate dominant modes. The algorithm has been tested on 24
dataset of coronal view with great success and validated by a robust index with highly
encouraging outcome. For validation study, some slices were selected such that the
M. Tech. (CE)
Page 17
2014-15
entire range of the image volume from anterior to posterior was covered. A domain
expert traced the target manually on each slice and it was compared with the
automatically drawn head and brain contour using the similarity index [21].
[D.K. Somwanshi, et-al, 2011] has proposed a method that provides complete
diagnosis of the internal structure of the body and even detects the smallest
abnormalities through MRI scanned images. The method introduced was the image
processing based method that performed the texture feature of MRI image and
segmentation has been done. Texture analysis of the 5 different images of each case is
done by finding their contrast, correlation, energy, homogeneity and entropy, and then
their range is obtained. In order to detect the presence of the abnormality in the image,
its texture analysis of both the normal and abnormal images were done. On the basis of
the values of abnormal image, the range was calculated and further the texture feature of
normal image was compared and the feature lying outside the range finally detects the
abnormality in the biomedical image [26].
[V. Salai Selvam, et-al, 2011] has proposed a Scalp EEG with modified Wavelet-ICA
and Multi-Layer Feed Forward Neural Network for Brain Tumor Detection. Use of
scalp EEG for the diagnosis of various cerebral disorders is progressively increased.
The chosen EEG epoch belonged to a brain tumor case was considered as positive and
that it belonged to a normal case as negative. Then the four possibilities of the network
outcomes were True Positive (TP) if the network decided that a chosen EEG belonged
to a brain tumor case when it actually did, True Negative (TN) if the network decided
that a chosen EEG belonged to a normal case when it actually did, False Positive (FP) if
the network decided that a chosen EEG belonged to a brain tumor case when it actually
belonged to a normal case and False Negative (FN) if the network decided that a chosen
EEG belonged to a normal case when it actually belonged to a brain tumor case. From
the ROC it was clear that the performance of the proposed method in detecting the brain
tumor using the scalp EEG was very much encouraging [22].
[V.P. Gladis, et-al, 2011] has proposed method to segment a human brain of MRI
image in which tumor detection and characterization were considered using HSOM and
Wavelet packets feature spaces. In the first phase, the MRI brain image was acquired
from patients database, In that film artifact and noise were removed, and Hierarchical
Self Organizing Map (HSOM) was applied for image segmentation. The HSOM was the
M. Tech. (CE)
Page 18
2014-15
extension of the conventional self-organizing map used to classify the image row by
row. In the second phase, the feature of the MRI image was extracted first. Therefore,
the proposed method has the potential to non-invasively determine the type of the brain
tumor [9].
[M. Usman Akram, et-al, 2011] has proposed a method for automatic brain tumor
diagnostic system from MRI images. The system consists of three stages to detect and
segment a brain tumor. In the first stage, MRI image of brain was acquired and
preprocessing was done to remove the noise and to sharpen the image. In the second
stage, global threshold segmentation was done on the sharpened image to segment the
brain tumor. In the third stage, the segmented image was post processed by
morphological operations and tumor masking in order to remove the false segmented
pixels. The proposed method enhanced the MRI image and segments the tumor using
global thresholding. False segmented pixels were removed using morphological
operations. The proposed method was invariant in terms of size, shape and intensity of
brain tumor. Experimental results show that the method performed provided the 92%
accuracy in enhancing, segmenting and extracting the brain tumor from MRI images
[14].
[Wang Yang, et-al, 2011] has proposed a Contour-based and Region Based methods
for segmenting brain tumors in 3D magnetic resonance images. This method was
applicable to different types of tumor. It takes the advantage of Fuzzy Classification for
automating the algorithm and the good quality segmentation result of deformable
models to improve the segmentation. This was achieved by IKFCM classification
method, morphological operation and a parametric 3D deformable model First the brain
was segmented using a new approach, robust to the presence of Tumors. Then a Tumor
detection was performed, based on improved Fuzzy classification. Applications on
several data sets with different Tumor sizes and different locations show that the
method works automatically with high quality of segmentation and were robust to interindividual variability for all types of fully enhancing Tumors [29].
[K. S. Angel Viji, et-al, 2011] has developed a brain tumor segmentation method and
validated using MRI Data. In Preprocessing and Enhancement stage, medical image was
converted into standard formatted image. Segmentation subdivides an image into its
constituent regions or objects. This method can segment a tumor provided that the
M. Tech. (CE)
Page 19
2014-15
desired parameters are set properly. The author focuses on the computation of steepest
slope lines, and produces zero width watersheds. Also, this method has a high degree of
locality that it was suitable for parallel implementation. In this, after a manual
segmentation procedure the tumor identification, the investigations has been made for
the potential use of MRI data for improving brain tumor shape approximation and 2D &
3D visualization for surgical planning and assessing tumor [18].
[K.K. Sharma, et-al, 2012] has proposed a threshold selection technique for color
image segmentation problem, which approximately preserves the colors in different
segments. The threshold values obtained are dependent on the particular definition of
the entropy chosen, which in turn affects the segmentation results. In the proposed
approach, threshold selection in each of the three component (RGB) images is done on
the basis of different entropy measures. It was further observed that the segmentation
results obtained using Havrda-Charvat entropy measures were better than other entropy
measures in the sense of preservation of colors in different segments [12].
[Rupsa Bhattacharjee, et-al, 2012]has proposed a novel algorithm to feature out tumor
from diseased brain Magnetic Resonance (MR) images. In this work, based on a study
of quality parameter comparison of two filters, adaptive median filter was selected for
de-noising the images. Image slicing and identification of significant planes were done.
Logical operations were applied on selected slices to obtain the processed image
showing the tumor region. A novel image reconstruction algorithm was developed
based on the application of Principal Components Analysis (PCA). This reconstruction
algorithm was applied on original raw images as well as on the processed images.
Results of this work confirm the sole efficiency of the developed image processing
algorithm to detect brain tumor [9].
[Azian Azamimi Abdullah, et-al, 2012] has proposed a brain tumor detection method
based on cellular neural networks (CNNs). CNN was widely used for image processing,
pattern, target classification, motion detection and signal processing. To examine the
location of tumor in the brain, Magnetic Resonance Imaging (MRI) was used.
Radiologists evaluated the grey scale MRI images. This procedure was really time and
energy consuming. To overcome this problem, an automated detection method for brain
tumor using CNN was developed. The output of the images was the image segmentation
based on black and white. Black represented as normal area and the white area
M. Tech. (CE)
Page 20
2014-15
represented as tumor area. By using the template in the CNN simulator, output of the
desired image could be performed. Therefore, many templates were combined in order
to obtain an accurate result that will help radiologists detecting the tumor in brain
images easily [3].
[Sahar Ghanavati, et-al, 2012] has proposed an automatic tumor detection algorithm
using multi-modal MRI. A multi-modality framework for automatic tumor detection
was presented, fusing different Magnetic Resonance Imaging modalities including T1weighted, T2-weighted and T1 with gadolinium contrast agent. The intensity, shape
deformation, symmetry, and texture features were extracted from each image. The
AdaBoost classifier was used to select the most discriminative features and to segment
the tumor region. Multi-modal MRI Images with simulated tumor was used as the
ground truth for training and validation of the detection method. The results on
simulated and patient MRI show 100% successful Tumor detection with average
accuracy of 90.11% [10].
[Sarah Parisot, et-al, 2012] has proposed a novel approach for detection segmentation
and characterization of brain tumors. In this the author combines an image-based
detections schema with identification of the tumors corresponding preferential location,
which was associated to a specific spatial behavior. This method exploits prior
knowledge in the form of a sparse graph representing the expected spatial positions of
tumor classes. Such information was coupled with image-based classification
techniques along with spatial smoothness constraints towards producing a reliable
detection map within a unified graphical model formulation. Towards optimal use of
prior knowledge, a two-layer interconnected graph was considered with one layer
corresponding to the low-grade glioma type (characterization) and the second layer to
voxel-based decisions of tumor presence. Efficient linear programming both in terms of
performance as well as in terms of computational load was considered to recover the
lowest potential of the objective function. The outcome of the method refers to both
tumor segmentation as well as their characterization [16].
[Yao Tien Chen, et-al, 2012] has proposed an approach integrating 3D Bayesian level
set method with volume rendering for brain tumor and tissue segmentation and
rendering. A prior probability estimation of the tumor and tissue was incorporated into
3D Bayesian level set method for 3D segmentation. The 3D Bayesian level set method
M. Tech. (CE)
Page 21
2014-15
was then used to continuously segment the 3D targets from a series of brain images. To
facilitate 3D volume visualization of medical-image dataset, ray casting was conducted
to render the targets and construct the surface of the targets. Experiment results were
finally reported in terms of segmenting, rendering, and surface reconstructing of the
tumor, tissue, and whole brain [12].
[Natarajan P, et-al, 2012] has proposed a morphological processing technique that has
proved miraculously helpful in various image extraction and filtering techniques. The
morphological operators changed the structuring elements of the image according to
their use. Some operators like open, spur, dilate and close had proved helpful in
extracting the brain tumor from the MRI brain images. Pre-processing of the MRI was
done using gray scaling, histogram equalization and filtering techniques. Threshold
segmentation was used to work on the desired region of the image. Thus by applying the
image subtraction could get the final brain tumor image [13].
[Vijayshree Gautam, et-al, 2013] has proposed an approach for threshold selection
purpose in color image segmentation problems. The threshold values obtained were
dependent on the particular definition of the entropy chosen, which in turn affects the
segmentation results. It was further observed that the segmentation results obtained
using Havrda-Charvat entropy measures are better than other entropy measures in the
sense of preservation of colors in different segments. Simulation results performed in
MATLAB. The entropy function versus individual RGB component level plot is
obtained for Shannon and non-Shannon entropy measures. A typical plot for entropy
function using Havrda-Charvat entropy measure was done [18].
[Pavel Dvorak, et-al, 2013] has proposed a method for the detection of images
containing an abnormality caused by tumor. The goal was to determine whether the
MRI Image of a brain contains a tumor. The proposed method works with T2-weighted
magnetic resonance images, where the head was vertically aligned. The detection was
based on checking the left-right symmetry of the brain, which was the assumption for
healthy brain. The algorithm was tested by five-fold cross validation technique on 72
images of brain containing Tumors and 131 images of healthy brain. The proposed
method reached the true positive rate of 91.16% [7].
[T.Rajesh, et-al, 2013] has proposed an automated system for classification of MRI
brain images with different pathological condition. Many cancer forms can only be
M. Tech. (CE)
Page 22
2014-15
diagnosed after a sample of suspicious tissue has been removed and tested. Pathologists
view pathologic tissues, typical1y with microscopes, to determine the degree of
normalcy versus disease. This process was time consuming and fatiguing. The system
described here classified the abnormality into benign or malignant in an automated
fashion. The author used conceptual1y simple classification method using Feed Forward
Neural Network. Texture features were calculated using Rough set theory. The
proposed system effectively classified the abnormality of brain tumors [15].
[Atiq Islam, et-al, 2013] has proposed a novel multifractal (multi-FD) feature
extraction and supervised classification techniques for improved brain tumor detection
and segmentation. The multi-FD feature characterizes intricate tumor tissue texture in
brain MRI as a spatially varying multifractal process in brain MRI. On the other hand,
the proposed modified AdaBoost algorithm considers wide variability in texture features
across hundreds of multiple-patient MRI slices for improved tumor and non-tumor
tissue classification. Experimental results with 14 patients involving 309 MRI slices
confirm the efficacy of novel multi-FD feature and modified AdaBoost classifier for
automatic patient independent tumor segmentation. In addition, comparison with other
state-of-the-art brain tumor segmentation techniques with publicly available low-grade
glioma inBRATS2012 dataset presented that the method outperform other methods for
most of the patients. The computation complexity of multi-FD feature was linear and
increased with slice resolution (number of pixel), block size, and the number of wavelet
levels [11].
[J.Vijay, et-al, 2013] has proposed a combine method of segmentation and K-means
clustering that described an efficient method for automatic brain tumor segmentation for
the extraction of tumor tissues from MRI Images. The main argument of the proposed
modifications is on the reduction of intensive distance computation that takes place at
each run (iteration) of K-means algorithm between each data point and all cluster
centers. In this method segmentation was carried out using K-means clustering
algorithm for better performance. This enhances the tumor boundaries more and was
very fast when compared to many other clustering algorithms. The proposed work also
reduces the computational complexity and also provides an accurate method of
extracting the Region of Interest (ROI).The proposed technique produced appreciative
results [27].
M. Tech. (CE)
Page 23
2014-15
[Woo Kyung Moon, et-al, 2013] has proposed a computer-aided detection (CAD)
system based on multi-scale blob detection for analyzing ABUS images. After speckle
noise reduction, Hessian analysis with multi-scale blob detection was adopted to detect
the lesions by using blobness measurements of ABUS images. Tumor candidate
selection was then applied to remove the redundant non-tumors from the tumor
candidates. The tumor likelihoods of the remaining tumor candidates were estimated
using a logistic regression model with blobness, internal echo, and morphology features.
Finally, the tumor candidates with tumor likelihoods higher than a specific threshold
were considered to be tumors [7].
[Solmaz Abbasi, et-al, 2014] has proposed a method for 3D medical image
segmentation to detect brain tumor in MRI images by combining Clustering and
Classification methods to decrease the complexity of time and memory. The author
presented a hybrid method in which clustering(for obtaining region of interest) and
classification (for region of interest- segmentation) were used. In the first phase, nonnegative matrix factorization with sparseness constraint method was used to separate the
region of interest from the image. In the second phase, the classification of the region of
interest was performed. This method had achieved a fast speed for segmentation of MRI
3D images and evaluated with criteria of Dice's and Jacquard's coefficient on the brain
tumor from magnetic resonance image obtained from the Brats2013 database [2].
[Akram Rashid, et-al, 2014] has proposed a method from Electroencephalogram that
eliminate the noise of power line, noise of human body muscles, noise of human lungs
and noise of the base line. In this the basic important algorithms used were Kalman
filter. The Electroencephalogram signals were highly contaminated with various
artifacts both from subject and from equipment interferences. For efficient detection of
tumor artifacts exist in the electroencephalogram signal were removed using analogue
filtering. After than the Fast Independent Component Analysis algorithm was used to
separate the noise and get the features which were buried in the extended band of noise.
For problem solution a unique Fast Independent Component Analysis filter was being
proposed [12].
[Parveen, et-al, 2015] has proposed a new hybrid technique based on the support vector
machine (SVM) and fuzzy c-means for brain tumor classification. In this technique the
image was enhanced using enhancement technique such as contrast improvement, and
M. Tech. (CE)
Page 24
2014-15
mid-range stretch. The brain MRI images were classified using SVM techniques which
were widely used for data analyzing and pattern recognizing. It creates a hyper plane in
between data sets to indicate which class it belongs to. Double thresholding and
morphological operations were used for skull striping. Fuzzy c-means (FCM) clustering
was used for the segmentation of the image to detect the suspicious region in brain MRI
image. The main objective of this work was to develop a hybrid technique, which can
classify the brain MRI images successfully and efficiently via Fuzzy C- means and
support vector machine (SVM) [18].
[ Padmakant Dhage, et-al, 2015 ] has developed a technique to solve the problem for
tumor case of clinical MRI analysis. The proposed technique was based on Watershed
segmentation and it has been successfully tested on MRI image data. Developed
algorithm was used to know about the location and size of the tumor. The author
illustrates the ability of watershed segmentation to separate the abnormal tissue from the
normal surrounding tissue to get a real identification of involved and noninvolved area
that help the surgeon to distinguish the involved area precisely. At the end of the
process tumor was extracted from the MRI Image and its exact position and shape was
determined and various parameters like perimeter, eccentricity, entropy and centroid
was calculated. All parameters which were extracted using developed algorithm specify
the size and other dimensions of the tumor [8].
2.2.2 Common Findings obtained in the issue Brain Tumor Detection from MRI
Images:
In the contour deformable model with regional base technique, the performance
was insufficient to obtain the fine edge in the tumor.
The efficient system was presented to perform early detection of brain Tumors
from EEG Signals with the aid of Artificial Neural Networks.
The feed forward back propagation neural networks provide computationally

more efficient Detection Accuracy with 98.7654% once trained.
Feature selection algorithms were categorized into Exponential, Randomized

and Sequential algorithms to attain high accuracies.
The Genetic Algorithm based Global Search was used for selecting the best set
of features.
M. Tech. (CE)
Page 25
2014-15
The classification of normal and tumor regions was done using texture features
from gray level dependence method (SGLDM) and wavelet features.
To improve the quality of images and limit the risk of distinct regions fusion in
the segmentation phase an enhancement process was applied on brain images to
increase the contrast in MRI images and to extract the suspicious regions or
tumors.
The segmentation and the localization of suspicious regions were performed by

applying the wavelet transforms.
The HSOM was the extension of the conventional self-organizing map used to
classify the image row by row.
A mathematical morphology was adopted to increase the contrast in MRI

images.
The Hybrid algorithm used Contour-based and Region based Methods to

segment Brain Tumors in 3D MRI Images.
The HSOM and Wavelet packets feature spaces method have the potential to
non-invasively determine the type of the brain tumor
Multi-modal MRI Images with simulated tumor were used as the ground truth
for training and validation of the detection method.
The performance of clustering algorithm for image segmentation was highly

sensitive to features that were used and the types of objects in the image and
hence generalization of this technique was difficult.
2.3 Various solution approaches in the issue Brain Tumor Extraction

from MRI Images
The Conceptual explanation of various solution approach and algorithms used are
summarized in the table below. This section would include, the issue-wise solution
approaches, methodologies used by the researchers, with results obtained. Based on
Shannon Algorithm different authors have used different data that is presented below:
M. Tech. (CE)
Page 26
2014-15
Table 2.3: Categorical Review of Research Paper

Author
Solution
Approach
Input
Parameters
Accuracy
(%)
[C.C. Leung, et-al,

2003]
Generalised
Fuzzy Operator
(GFO)
MRI Images
98
[Ahmed Kharrat, etal, 2009]
Mathematic
Morphology
MRI Images
87
[M. Stella Atkins,etal,1998]
Automatic
intracranial
boundary
detection
algorithm
Performance Parameters
Comparison with CDM and modified GFO(mGFO)
Real edge
CDM
mGFO
Total no. of
1749
2443
1783
pixels within
tumor
Number of
1749
+694 (41.09%)
+34 (2.01%)
mismatch pixels
(1689)
out of tumor
Number of
1749
-60 (100%)
-4 (6.67%)
mismatch pixels
(60)
inner of tumor
Mathematic
MRI Image
CLAHE
Beghadi
Morphology
Abnormal Image
MRI data
sets
96
Watershed
Segmentation
[AzianAzamimi
Abdullah, et-al,
2012]
Combined
Clustering and
classification
Mechanism
Cellular Neural
Networks
(CNN)
[SaharGhanavati, etal, 2012]
Multi-Modality
Framework
[Wang Yang, etal,2011]
Hybrid
Algorithm
[SolmazAbbasi, et-al,
2014]
MRI Images
3D Medical
Image
97.3
N/A
1
2
3
4
Dice
Complte
Tumor
0.8497
MRI Images
T1 and T2
weighted
Magnetic
Resonance
Imaging
3D
Magnetic
0.9997
Parameter
MSE
PSNR
Contrast
Corelation
Dice
Dice
Tumor
Enhancing
Core
Tumor
0.7902
N/A
90.11
0.8172
1.5073
Comparision of Manual and Automatic Brain Segmentation

Similarity
Slice No.
Manual Area
Auto Area
Index
4
8912
10269
0.925
7
20264
22472
0.929
8
22035
23918
0.954
23
20109
20341
0.980
24
15909
17201
0.958
S.No.
[PadmakantDhage,
et-al, 2015]
Results
0.7941
Median Filter
Bilateral Filter
2.98
43.3
0.10
0.99
4.05
42.0
0.01
0.99
Jaccard
Complete
tumor
Jaccard
Tumor
Core
Jaccard
Enhancing
Tumor
0.7467
0.6678
0.6884
The extraction of the tumor was accomplished

manually from segmentation result and it was
observed that Mathematical Morphology was better.
The algorithm has proven effective on clinical and

research MRI data sets acquired from several
different scanners
Accurately remove the noise from MRI Images

without disturbing the edges.
This method decreases the computational complexity in

time and memory.
The output of the image detected in shorter time and based
on black and white. Black portion represented as normal
area and white are represented as tumor area.
N/A
Cases
Accuracy
Specificity
Case1
90.11%
92.26%
Case2
89.12%
True Detection
Tp = NTP/NM*100%
91.19%
M. Tech. (CE)
Accurately search the boundary of the Brain Tumor.
The better segmentation at the edges of tumor was obtained.
The average computation time for detection of tumor was

reduced.
Page 27
(Contour based
and region
based)
Resonance
Images
N/A
Tumor Type
FE1
FE1
FE2
FE2
[Satish Chandra, etal,2009]
Clustering
Algorithm
based on
PSO
110
abnormal
and 62
normal axial
MRI images
[T.Logeswari, et-al,
2010]
Hierarchical
Self
Organizing
Map
MRI Data
N/A
[M. Murugesan, etal, 2009]
Artificial
Neural
Network
325 samples
of EEG
data.
98.7654
[V. SalaiSelvam, etal, 2011]
Scalp EEG
withModified
Wavelet-ICA
MRI and
CT-SCAN
Data
N/A
[RupsaBhattacharjee,
et-al, 2012]
PCA Based
Reconstruction
92.41
Global
thresholding
100 MRI
Images
1.92
7.63
9.57
4.40
8242.5
8.9702
0.0742
1590.2
16.1163
0.4306
MSE
1271.4
424.1774
PSNR
17.0879
21.8553
UIQI
Parameter
Standard Deviation
Accurately Segmented
0.1894
N/A
97
82
93
93
98
MSE
PSNR
UIQI
Guassian
Noise
[M. Usman Akram,

et-al, 2011]
FPCM
IKFCM
FPCM
IKFCM
FP
%
Results for various Algorithm

Classifier
Precision
Recall
Accuracy
Proposed PSO
92.76
96.24
94.42
Based algorithm
SVM (Polynomial
93.33
95.28
92.71
kernel)
AdaBoost
90.25
91.66
89.31
Number
neighborhood
Winning
of seg
Exe-time
weight
pixel
neuron
pixel
3x3
209
795
13.76
14
5x5
201
1073
14.96
8
7x7
194
1285
15.20
15
9x9
186
1594
11.05
23
11x11
177
1881
11.53
32
Detection Accuracy
325 Samples of EEG
Normal
94.4785
Data
Abnormal
98.7654
0.930
Sensitivity or TPR
0.106
FPR
0.918
Accuracy
0.894
Specificity or TNR
Filters Comparision
Noise Details
Parameter
Bilaterl Filter
Adaptive
Noise Density
MRI Images
False Detection
Fp=NFP/NA*100%
TP
Method
%
Poorly Segmented
M. Tech. (CE)
2014-15
Performance is improved in terms of execution time and

tumor pixel detected.
The target area is

segmented and Brain Tumor detection done accurately
The trained Feed forward back propagation neural network

has detected the presence of brain tumor in the test signal.
Improved the detection rate of the Tumor.
Abnormal area of the brain is detected and pixel wise

intensity is calculated.
0.2520
Value
0.0013
97%
3%
The proposed method performs well in enhancing,

segmenting and extracting the brain tumor from MR
images.
Page 28
[Woo Kyung Moon,

et-al, 2013]
[Natarajan P, et-al,
2012]
[T.Rajesh, et-al,
2013]
[M. Sasikala, et-al,

2005]
Multi-Scale
Blob Detection
Algorithm
Threshold
operation
ABUS
image
MRI Brain
Images
Feed Forward
Neural
Network
MRI Brain
Images
Artificial
Neural
Network
19 scans of
glioblastoma
multiforme
and 18 scans
containing
normal brain
images
97
Feature Set
Blobness
Internal Echo
Morphology
Blobness and
Internal Echo
Blobness and
Morphology
Internal Echo and
Morphology
All features with
IOF-CV
All features with
LOO-CV
Evaluation
Metrices
(TN)True Negative
(FP)False Positive
(TP)True Positive
(FN)False Negative
Specificity
Sensitivity
Accuracy
Feature Selection
Algorithm
SFS
97.3%
SBS
Plus 1- take away r
method
SFFS
FOM
0.28
0.23
0.16
39.81
0.33
61.08
0.36
37.81
0.43
17.33
0.47
17.45
0.47
RGB
1. For black
R = G = B = 00000000
2. For white
R = G =B = 11111111
N/A
90
False Positive
113.26
42.04
65.16
Training Dataset
9
1
10
0
0.9
1
0.95
Feature Set
ASM, IDM, SVAR,
SENT
SVAR,ENT,
ENERGY, MEAN
ASM, SVAR, IMC,
ENERGY,
ASM, SVAR, SENT,
DENT
M. Tech. (CE)
2014-15
Successfully detected the abnormalities with tumor region.
Tumor is identified from the MRI Image accurately.
Proposed Testing
Dataset
10
0
8
1
1
0.88
0.9
Classifier
Accuracy
Suspicious tissues are tested and many cancer forms are

diagnosed.
97.3%
97.3%
The normal and tumor images are classified with an

accuracy of 97.3%
97.3%
97.3%
Page 29
2014-15
Table 2.4: Comparison Table of different Entropy
Author
Solution
Approach
Data used
Performance Measures
(Threshold Value)
[Shruti
Mathur, et-al,
2014]
Shannon and nonShannon Entropy
jpg image
[Vijayshree
Gautam, et-al,
2013]
Shannon and Nonshannon Entropy
Color Image
Entropy
Shannon
Kapur
Vajda
Renyi
Havrda
Entropy
R
124
[Ahmad Adel
Abu Shareha,
et-al, 2008]
Renyis entropy
Synthetic
Images
[Siddheswar
Roy, et-al,
2000]
Shannon Entropy
Three images
( balls,
molecule, rose)
[K.K. Sharma,
et-al, 2012]
Shannon and NonShannon Entropy
Ball Image
Algae.jpg
America.jpg
83
197
140
211
141
209
140
211
113
161
Component
G
B
42
55
Havrda-Charvat
Entropy
Shannon Entropy
17
Renyi Entropy
234
Image No. Threshold using the
original Entropy
1
T=187
2
T=155
3
T=55
4
T=154
5
T=153
6
T=133
7
T=178
8
T=157
Entropy
Entropy
Balls
Function
type
Shannon
Global
369.7350
Shannon
Local
296.1081
Shannon
Conditional 227.3324
Entropy
Kapur
HavrdaCharvat
Shannon
Kaju.jpg
10
26
5
122
83
Results
R
20
39
106
15
39
Threshold using
Textured Renyi
T=177
T=86
T=55
T=74
T=149
T= 133
T= 172
T= 154
Molecule Rose
194.3193
316.2784
324.0741
Component
G
23
Various individual RGB component

level plot is obtained for Shannon and
non-Shannon entropy measures.
The experimental results show that the

proposed
method
enhances
the
thresholding result and reduce the error
rate.
The experimental results was quite

satisfactory.
180.9309
204.4700
219.7089
B
71
117
138
141
64
80
198
M. Tech. (CE)
Havrda-Charvat entropy based image

segmentation technique is better for gray
images than any other entropy measures.
The segmentation results obtained using

Havrda-Charvat entropy measures were
better than any other entropy measures
Page 30
[P.K. Saboo,
et-al, 2006]
Havrda-Charvat
Entropy
Tif images
[Amar Pratap
Singh, et-al,
2009]
Shannon and NonShannon Entropy

Measures
Mammogram
Image Analysis
Society
Database
Baljit Singh
Khehra,etal,2011
Shannon and nonshannon Entropy
mini-MIAS
database
(Mammogram
Image Analysis
Society
database
Renyi
Vajda
Test Image
Camaraman.tif
Kids.tif
Rice.tif
Tire.tif
Bonemarr.tif
Boats.tif
Bridge.tif
Lena.tif
Images
S
Sample
Mam1
Mam2
Mam3
Mam4
Mam5
Mam6
Test Image
mdb218
mdb236
mdb238
mdb219
mdb222
mdb227
mdb240
mdb245
mdb248
mdb253
216
10
4.76
5.17
5.24
5.16
4.84
5.10
tS
180
186
171
185
186
184
198
177
177
191
R
=0.5
HC
=0.7
6.92
6.90
7.01
6.71
6.78
6.79
tR
191
200
179
194
193
193
210
198
192
203
8.79
9.05
9.34
8.63
8.51
8.71
tHC
194
201
178
198
194
194
210
200
193
205
M. Tech. (CE)
215
21
T(0.3)
99
32
182
111
122
102
125
122
K
=0.5
=0.7
7.99
7.79
7.88
7.57
7.81
7.72
tk
193
201
177
198
194
194
210
199
194
205
2014-15
73
72
The images were segmented perfectly
and a histogram was created using the
gray value.
Results have demonstrated that Havrda

& Charvat entropy based feature for
classifying normal and abnormal
mammogram images.
The result show that it is useful for

radiologists to find suspicious region
in mammogram.
Page 31
2014-15
2.4 Strengths and weaknesses of research found in the issue Brain Tumor
Extraction from MRI Images using Entropy Measure
After, the review of 32 research papers under the area of Brain Tumor Extraction from
MRI Images, I have been able to find, the strengths and weaknesses of various solution
approaches used to solve the issues discussed in previous chapters. This section would
enlist the strengths and weaknesses of the various methods and algorithms used.
2.4.1Strengths in the area of Brain Tumor Detection from MRI Images:
After, the review of thirty one research papers under the area of Brain Tumor
Extraction from MRI Images strengths and weaknesses of various solution approaches
used could be determined. This section mainly enlists the strengths of the various
methods and algorithms used.
The watershed transformation algorithm extracts shape and form related

information from images precisely [8].
The Vajda entropy measures provide the advantage of faster calculations over
Kapurs entropy measure [17].
Havrda-Charvat entropy based image segmentation technique was better for

gray images than any other entropy measures [12].
The major advantage of split and merge technique that it guaranteed connected
regions [21].
The Mean Shift algorithm provides a robust feature space analysis approach
which can be applied to discontinuity preserving smoothing and image
segmentation problems [6].
The Ncut method was applied directly to the image pixels, which were typically
of very large size [4].
The Ncut method was applied to perform globally optimized clustering and get
the final segmentation result [13].
The Mean Shift algorithm can significantly reduce the number of basic image
entities, and due to the good discontinuity preserving filtering characteristics, the
salient features of the overall image were retained [1].
Edge detection algorithm was used for object detection which serves various
applications like medical image processing, biometrics etc [23].
M. Tech. (CE)
Page 32
2014-15
Thresholding was computationally inexpensive and fast, it was the oldest

segmentation method and was still widely used in simple applications [27].
Global thresholding method selects only one threshold value for the entire image
[3].
Edge preserving denoising was of great concern in medical images. Denoising

was performed to segmentize the image for better diagnosis [11].
2.4.2Weaknesses in the area of Brain Tumor Detection from MRI Images:
By applying the Ncut algorithm the performance and stability of the partitioning
highly depends on the choice of the parameters [19].
The drawbacks of the split and merge technique were, the results depend on the
position and orientation of the image, leads to blocky final segmentation and
regular division leads to over segmentation (more regions) by splitting [12].
If seeded region growing method was used then noise in the image can cause the
seeds to be poorly placed [8].
Local thresholding selects different threshold values for different regions [23].
The performance of clustering algorithm for image segmentation was highly

sensitive to features used and types of objects in the image and hence
generalization of this technique is difficult [15].
Clustering technique doesnt guarantee continuous areas in the image, even if it

does edges of these areas tend to be uneven, this was the major drawback [29].
2.5 Gaps in the published work

The Researcher had presented many techniques and methodologies to segment the MRI
images and to detect the Brain Tumor. All the Researchers had shown some
improvement with some gaps in their work. Those are:
A very few approaches had been made for Early Detection and classification of
Brain Tumors.
There are various approaches that have been proposed to deal with the task of
segmenting and detecting the brain tumors in MRI Images but the performance
of these approaches usually depends on the qualitative amount of information.
M. Tech. (CE)
Page 33
2014-15
Most of the research work was directed towards the speed of computation that
was no longer an issue nowadays. The improvement of information obtained
from images and perfecting the process of segmentation to get an accurate
picture of the brain tumor must be given importance.
During the acquisition of medical images, there were possibilities that the
medical image one gets might be degraded. Very few works has been done in
order to reduce this problem.
2.5 Problem Statement

Brain Tumor was one of the frequent and leading causes of mortality, especially in
developed countries. Though brain tumor leads to death, early detection can increase the
survival rate. In this dissertation work the main emphasis laid on to design an approach,
which was a detection technique so that the system effectively detects and diagnose the
tumor in their early stage.
Thus the final title of the work chosen to achieve the main aim of early detection of
tumors in MRI Images is An efficient Brain Tumor Extraction from MRI Images
using Entropy Measures.
2.6 Final Objectives

1. To finalize the MRI images for experimentation.
2. To perform the global thresholding image Segmentation on MRI Images
3. To calculate the entropy and to check the threshold of each image in order to find
out the tumor affected region.
4. To compare and analyze the performance of the Shannon and non-Shannon entropy
measures.
So this chapter presents, Performance Evaluation Criteria used by various researchers,
details of Software used by researchers, Methodologies used by Researchers. Next
Chapter will provide the theoretical aspect of targeted work, details of Input/output
parameters that will be used during experimentations and details of softwares which
have been selected to carry out experimentation.
M. Tech. (CE)
Page 34
2014-15
Chapter 3: Theoretical Aspects of Brain Tumor Extraction from MRI

Images
In previous chapter we have discussed literature review of different problems and
solutions in brain tumor detection. This dissertation focuses on accurate detection of
brain tumor, image segmentation, to minimize diagnostic errors, to improve the
accuracy by using Computer aided diagnostic. Different types of tumor detection
methods and some other techniques in order to get an accurate picture of brain tumor
from images obtained through the slice orientation and perfecting the process of
segmentation are discussed in this chapter.
In day-to-day life, new technologies are emerging in the field of Image processing,
especially in the domain of segmentation. Segmentation is the most important part in
image processing Fence off an entire image into several parts which is something more
meaningful and easier for further process. Segmentation may also depend on various
features that are contained in the image. It may be either color or texture. Before
denoising an image, it is segmented to recover the original image. The main motto of
segmentation is to reduce the information for easy analysis. Segmentation is also useful
in Image Analysis and Image Compression.
3.1 Classification of Image Segmentation

There are several existing techniques which are used for image segmentation. These all
techniques have their own importance.
Threshold Methods
Edge Based Methods
Region Based Methods
Image Segmentation
Methods
Clustering Based Methods

Watershed Based Methods
PDE Based Methods
ANN Based Methods
Fig. 3.1 Image Segmentation Techniques
The popular techniques used for image segmentation are: thresholding method, edge
M. Tech. (CE)
Page 35
2014-15
detection based techniques, region based techniques, clustering based techniques,

watershed based techniques, partial differential equation based and artificial neural
network based techniques. These all techniques are different from each other with
respect to the method used by these for segmentation.
3.1.1 Thresholding Method
Thresholding methods are the simplest methods for image segmentation. These methods
divide the image pixels with respect to their intensity level. These methods are used
over images having lighter objects than background. The selection of these methods can
be manual or automatic i.e. can be based on prior knowledge or information of image
features. There are basically three types of thresholding.
1. Global Thresholding: This is done by using any appropriate threshold value/T. This
value of T will be constant for whole image.
2. Variable Thresholding: In this type of thresholding, the value of T can vary over the
image. This can further be of two types
Local Threshold: In this the value of T depends upon the neighborhood of x and y.
Adaptive Threshold: The value of T is a function of x and y.
3. Multiple Thresholding: In this type of thresholding, there are multiple threshold

values like T0 and T1. The values of thresholds can be computed with the help of the
peaks of the image histograms. Simple algorithms can also be generated to compute
these.
1.1.2 Edge Based Segmentation Method
The edge detection techniques are well developed techniques of image processing on
their own. The edge based segmentation methods are based on the rapid change of
intensity value in an image because a single intensity value does not provide good
information about edges. Edge detection techniques locate the edges where either the
first derivative of intensity is greater than a particular threshold or the second derivative
has zero crossings. In edge based segmentation methods, first of all the edges are
detected and then are connected together to form the object boundaries to segment the
required regions. The basic two edge based segmentation methods are: Gray histograms
and Gradient based methods. To detect the edges one of the basic edge detection
techniques like sobel operator, canny operator and Roberts operator etc can be used.
Result of these methods is basically a binary image. These are the structural techniques
M. Tech. (CE)
Page 36
2014-15
based on discontinuity detection.

3.1.3 Region Based Segmentation Method
The region based segmentation methods are the methods that segments the image into
various regions having similar characteristics. There are two basic techniques based on
this method
1. Region growing methods: The region growing based segmentation methods are the
methods that segments the image into various regions based on the growing of seeds
(initial pixels). These seeds can be selected manually (based on prior knowledge) or
automatically (based on particular application). Then the growing of seeds is controlled
by connectivity between pixels and with the help of the prior knowledge of problem,
this can be stopped.
2. Region splitting and merging methods: The region splitting and merging based
segmentation methods uses two basic techniques i.e. splitting and merging for
segmenting an image into various regions. Splitting stands for iteratively dividing an
image into regions having similar characteristics and merging contributes to combining
the adjacent similar regions.
3.1.4 Clustering Based Segmentation Method
The clustering based techniques are the techniques, which segment the image into
clusters having pixels with similar characteristics. Data clustering is the method that
divides the data elements into clusters such that elements in same cluster are more
similar to each other than others. There are two basic categories of clustering methods:
Hierarchical method and Partition based method. The hierarchical methods are based on
the concept of trees. In this the root of the tree represents the whole database and the
internal nodes represent the clusters. On the other side the partition based methods use
optimization methods iteratively to minimize an objective function. In between these
two methods there are various algorithms to find clusters. There are basic two types of
clustering.
1. Hard Clustering: Hard clustering is a simple clustering technique that divides the
image into set of clusters such that one pixel can only belong to only one cluster. In
other words it can be said that each pixel can belong to exactly one cluster. These
methods use membership functions having values either 1 or 0 i.e. one either certain
pixel can belong to particular cluster or not. An example of a hard clustering based
M. Tech. (CE)
Page 37
2014-15
technique is one k-means clustering based technique known as HCM. In this technique,
first of all the centers are computed then each pixel is assigned to nearest center. It
emphasizes on maximizing the intra cluster similarity and also minimizing the inter
cluster equality.
2. Soft clustering: The soft clustering is more natural type of clustering because in real
life exact division is not possible due to the presence of noise. Thus soft clustering
techniques are most useful for image segmentation in which division is not strict. The
example of such type of technique is fuzzy c-means clustering. In this technique pixels
are partitioned into clusters based on partial membership i.e. one pixel can belong to
more than one clusters and this degree of belonging is described by membership values.
This technique is more flexible than other techniques
3.1.5 Watershed Based Segmentation Methods
The watershed based methods uses the concept of topological interpretation. In this the
intensity represents the basins having hole in its minima from where the water spills.
When water reaches the border of basin the adjacent basins are merged together. To
maintain separation between basins dams are required and are the borders of region of
segmentation. These dams are constructed using dilation. The watershed methods
consider the gradient of image as topographic surface. The pixels having more gradient
are represented as boundaries which are continuous
3.1.6 Partial Differential Equation Based Segmentation Method
The partial differential equation based methods are the fast methods of segmentation.
These are appropriate for time critical applications. There are basic two PDE methods:
non-linear isotropic diffusion filter (used to enhance the edges) and convex nonquadratic variation restoration (used to remove noise). The results of the PDE method is
blurred edges and boundaries that can be shifted by using close operators. The fourth
order PDE method is used to reduce the noise from image and the second order PDE
method is used to better detect the edges and boundaries
3.1.7 Artificial Neural Network Based Segmentation Method
The artificial neural network based segmentation methods simulate the learning
strategies of human brain for the purpose of decision making. Now days this method is
mostly used for the segmentation of medical images. It is used to separate the required
image from background. A neural network is made of large number of connected nodes
M. Tech. (CE)
Page 38
2014-15
and each connection has a particular weight. This method is independent of PDE. In this
the problem is converted to issues which are solved using neural network. This method
has basic two steps: extracting features and segmentation by neural network.
3.1.8 Comparison of various Segmentation Techniques
Table shows a comparison between various segmentation techniques by specifying a
brief description of every method each with its advantages and disadvantages.
Table 3.1 Comparison of various Segmentation Techniques
Segmentation
Description
Advantages
Disadvantages
Technique
Thresholding
based on the
no need of previous
highly dependent on
Method
histogram peaks of
information,
peaks, spatial details
the image to find
simplest method
are not considered
particular threshold
values
Edge Based
based on
good for images
not suitable for
Method
discontinuity
having better
wrong detected or
detection
contrast between
too many edges
objects
Region Based
based on
more immune to
expensive method in
Method
partitioning image
noise, useful when
terms of time and
into homogeneous
it is easy to define
memory
regions
similarity criteria
Clustering
based on division
fuzzy uses partial
determining
Method
into homogeneous
membership
membership
clusters
therefore more
function is not easy
useful for real

problems
Watershed
based on
results are more
complex calculation
Method
topological
stable, detected
of gradients
interpretation
boundaries are
continuous
PDE Based
based on the
fastest method, best
M. Tech. (CE)
more computational
Page 39
Method
working of
for time critical
differential
applications
2014-15
complexity
equations
ANN Based
based on the
no need to write
more wastage of
Method
simulation of
complex programs
time in training
learning process for

decision making
3.2 Entropy Functions

Generally, entropy refers to disorder or uncertainty. Entropy is the expected value
(average) of the information contained in each message received. For color image
segmentation which approximately preserves the color in different segments, the
different entropy measures such as the Shannon, Renyi, Havrda- Charvat, Vajda and
Kapur Entropy measures are used for the simulation of results.
3.2.1Shannon Entropy
Shannon's entropy measure Hs(pm1,m2)provides an absolute limit on the best possible
lossless compression of a signal under certain constraints. It is defined as:
Hs(pm1,m2) =
m1 , m2
.. eq(1)
log p m1 ,m2
m1 m2
Where pm1,m2is the probability distribution associated with 2-D random variable. In this
paper, we have computed the values of pm1,m2 from the entries of the individual RGB
component level co-occurrence matrix of the given image and it is given by the relation
pm1,m2= Cm1,m2/(MN) where M, N represents the image dimensions along x and y
directions respectively. The entropy function for the purpose of the calculation of
threshold for image segmentation is then computed from the expression given as:
t
Entropy(t)= -
L 1
pm1 ,m2 log pm1 ,m2
m1 0 m2 t 1
L 1
m1 t 1 m2 0
m1 , m2
log p m1 ,m2
..eq(2)
Where, L represents the maximum number of Gray Level present in a particular image.
3.2.2Renyi Entropy
The Renyi entropy Hr(pm1,m2) which is a generalization of Shannon Entropy, is one of a
family of functional for quantifying the diversity, uncertainty or randomness of a
system. It is defined as:
M. Tech. (CE)
Page 40
2014-15
log pm1,m 2
m1 m 2
, 1, 0
H r pm1,m 2
1
And the corresponding entropy function is given by
t L 1
L 1 t
log ( p m1 ,m2 ) log ( p m1 ,m2 )

m1 0m2 t 1
m1 t 1 m2 0
Entropy(t) =
1
1
..eq (4)
3.2.3 Havrda-Charvat Entropy

The HavrdaCharvt entropy Hhc(pm1,m2) of degree , introduced by Havrda and
Charvt and later on modified by Daroczy, is often used in statistical physics and is
defined as follows:
H hc Pm1,m 2
m1, m 2
m1 m 2
1
1
..eq (5)
And the corresponding entropy function is given by

Entropy(t) =
1
21
t L 1
1 L 1 t
p m ,m 1 1
p m ,m 1
2 1 m t 1 m 0 1 2
1 m1 0m2 t 1 1 2
2
..eq (6)
3.2.4 Kapur Entropy

Kapurs entropy Hk(pm1,m2) of order and type defined as
H k Pm1,m 2
1
m1, m 2
m1 m 2
1
1 L 1 t
P
1
1 m1, m 2
2 m1t 11 m 20
..eq (7)
And the corresponding entropy is given by

L 1
t P 1
Pm1 ,m21
1
1
m ,m
1
1 2 1 1 2 1 21 1 ..eq (8)
Entropy(t) =
m101 m2 t 1 Pm1 , m2
m1 t 1 m2 0 Pm1 , m2
t
L 1
3.2.5 Vajda Entropy

Vajda entropy measure Hv(pm1,m2) is a special case of Kapurs entropy where =1 is
taken. It provides the advantage of faster calculations over Kapurs entropy measure and
is defined as
Pm1,m 2
H v Pm1,m 2 m1 m 2
1 (21 1) 1
Pm1,m 2

m1 m 2
M. Tech. (CE)
..eq (9)
Page 41
2014-15
And the corresponding entropy is given by
t L 1
L 1 t
p m1 ,m2
p m1 ,m2
m1 0m2 t 1
1
m1 t 1 m2 0
1
Entropy(t) = t L 1
1(2 1) L 1 t
1(21 1) 1
p m1 ,m2
p m1 ,m2
m1 0m2 t 1
m1 t 1 m2 0
..eq (10)
This chapter discussed the proposed solution approaches, details of software and various
methodologies used by Researchers. Next Chapter Contains system block diagram of
the target work, details of the Algorithm for Gray and color image segmentation, details
of Input/output parameters that will be used during experimentations and details of
softwares which have been selected to carry out experimentation.
M. Tech. (CE)
Page 42
2014-15
Chapter 4 Design Aspects of Proposed Work

The previous chapter focuses on various tumor detection techniques and was targeted to
evaluate the several entropy measures for threshold selection purpose in MRI Images.
This chapter presents, the details of the architectural design of the targeted work, details
of the input data to carry out experimentation and hardware and software which were
selected to carry out the experimentation to achieve the objectives.
4.1 System Design of the work

The design specification flow diagram provides the total conceptual diagram used for
experimenting.
Input Image
Preprocessing
{Resizing image for
permissible
computation}
Output Image
Feature Extraction
Segmentation
{Texture features
described using Cooccurrence Matrix}
{Separate image
information into
meaningful parts}
Diagnosis
Entropy
Calculation{Find the
{Detecting the
Tumor}
regional minima}
Figure 4.1 System Block Diagram
4.1.1 Input Image

MRI Images are taken as the input images in this dissertation work.
A high resolution MRI images are used for better visualization of the structure
of the brain.
In this work, the different dimension images of the MRI scan are used named as
the axial view, sagittal view and coronal view and is used for the diagnosis
purpose.
The different dimensions of the MRI Images are as follows
Dimension 1: Front to back of head
M. Tech. (CE)
Page 43
Dimension 2: Left to right of head
Dimension 3: Bottom to top of head
2014-15
MRI Scan images are chosen because MRI scan is less harmful than any other
diagnostic method.
4.1.2 Preprocessing
The MRI Images that is selected are of different sizes. To make all the mages in
a standard format the resizing of all the images are done for its permissible
computation.
All the MRI Images are converted into a dimension of 256*256 file format.
To perform the resizing of the image following command is used.
Imresize( a [256*256] )
This will create an image with 256 rows and 256 columns.
4.1.3 Feature Extraction:

A statistical method of examining texture that considers the spatial relationship
of pixels is the Gray-Level Co-occurrence Matrix (GLCM).
The GLCM characterize the texture of an image by calculating how often pairs
of pixel with specific values and in a specified spatial relationship occur in an
image, creating a GLCM, and then extracting statistical measures from this
matrix.
For each case, several co-occurrence matrices were calculated corresponding to
different MRI Images.
To create a gray-level co-occurrence matrix from image, glcm = graycomatrix(I)
is used.
4.1.4 Segmentation
Thresholding is one of the old, simple, and popular techniques for image
segmentation.
The global thresholding is applied to retrieve information from image.
The threshold of the matrix is determined as to make binary matrix
Matrix[i][j] = 1
if Matrix[i][j] <th
=0
Otherwise
[ i and j stands for ithrow and jth column ]
M. Tech. (CE)
; threshold ( th )
Page 44
2014-15
Now, no. of 1s are counted row wise and stored in an array of size n , [where n=
MN]
4.1.5 Entropy Calculation

Entropy is a statistical measure of randomness that can be used to characterize
the texture of the input image.
Here, Shannon and Non-Shannon Entropy measures are evaluated for threshold
selection.
The numbers of minima points are determined from the entropy function versus
gray level ( t) plot.
Entropy=1/information
4.1.6 Diagnosis
The tumor is diagnosed from the simulation results that were obtained from the
different Entropy calculation.
It was observed from the result that the Havrda-Charvat Entropy provides the
better result rather than the rest of the Entropies for quick detection of Tumor.
4.2 Details of data used: The details of data used to carry out experimentation are
as follows.
4.2.1 Input Images: The original MRI Image will be used as the input images. Table
4.1 shows properties of various MRI Image which are taken as input.
Table 4.1: Properties of Various Original Images
Properties
S. No.
MRI Images
Dimension(pixels)
Size(in KB)
grade1.png
445*470
48.0
grade1a.png
350*333
21.8
grade1b.png
352*414
42.7
grade2a.png
416*545
43.4
grade2b.png
462*307
25.9
grade2c.png
443*530
41.0
grade3a.png
225*257
16
grade3b.png
336*413
33.2
M. Tech. (CE)
Page 45
2014-15
grade3c.png
339*402
24.7
10
grade4a.png
227*292
15.6
11
grade4b.png
292*350
21.7
12
grade4c.png
235*301
20
4.3 Design and Implementation of the work carried out

4.3.1 Algorithm
Step 1: A M*N MRI Image is taken as input.
Step 2: Read image and resize it into 256*256 pixels.
Step3: The co-occurrence matrix Cm1,m2of the image is computed for each image.
Step 4: The probability distribution Pm1,m2= Cm1,m2/MN is then calculated from its cooccurrence matrixCm1,m2.
Step 5: The entropy function for each entropy definitions, are then calculated for each
image that is to be segmented.
Step 6: The number of minima points are then determined from the entropy function
versus gray level plot.
Step 7: The numbers of threshold points are determined from the entropy function and
the tumor detected region is calculated.
4.3.2 Process Flow Diagram
Process flow diagram of the work is discussed in this section. It involves the steps to
create an overview of the work carried out. The process flow diagram is shown below in
Fig. 4.2
Initialize the variable i,k
Read images from source directory
For i=1:1:k
Resize the original image 256*256 pixels
Create the Gray-Level Co-occurrence Matrix
GLCM=graycomatrix[I]
M. Tech. (CE)
Page 46
2014-15
Compute the probability distribution

Pm1,m2=Cm1,m2/MN
Calculate different Entropy
Plot Graph, Plot(x,y)
Compute the regional minima
Minimas=imregionalmin(a)
Convert the matrix into binary matrix
Plot the gray image
Tumor detected
Figure 4.2 Process flow diagram
This process flow provides the information about the common approach that is followed
in the work in order to carry out the result. First the image is taken as input image and
its probability matrix was created in order to convert the images in matrix form. By
using the value from matrix its different entropies were calculated in order to detect the
Tumors from MRI Images. After scanning the images from all entropies it was found
that Havrda-Charvat produces the best result from other entropies.
M. Tech. (CE)
Page 47
2014-15
4.3.2.1Flow chart for Shannon Entropy

Start
E(t)=0
K(t)=0
q(t)=0
For t=1:254
For m1=1:t+1
For m2=t+2:256
For m1=t+2:256
e(t)=e(t)+p(m1,m2)*
log(p(m1,m2))
For m2=1:t+1
K(t)=k(t)+(p(m1,m2))*l
og(P(m1,m2))
Symbols
Meaning
t
L 1
e(t)
m1 0 m2 t 1
ae(t)=-k(t)-e(t)
L 1
m1 , m2
k(t)
m1 t 1 m2 0
n=1:1:255
log p m1 ,m2
m1 , m2
log pm1 ,m2
ae(t)
-k(t)-e(t)
Maximum no. of gray level
0,1,2.L-2
m1,m2
0,1,2.L-1
Plot(n,ae)
End
Figure 4.3 Flow chart for Shannon Entropy
M. Tech. (CE)
Page 48
2014-15
4.3.2.2Flow chart for Renyi Entropy

Start
l=256, hh=2
sum=0, sum1=0
For t=1:254
For m1=1:1:t
For m2=t+1: 1: l -1
For m1=t+1:1:l-1
sum=sum+(p(m1,m2)^hh)
For m2=1:1:t
gg=log(sum)
gg1=log(sum1)
gg2=gg/(1-hh)
gg3=gg1/(1-hh)
Symbols
sum
t L 1
( p m1 ,m2 )
m1 0m2 t 1
sum1
L 1 t
( p m1 ,m2 )
m1 t 1 m2 o
e(t)=gg2+gg3
n=1:1:255
Plot(n,ae)
Meaning
gg
log(sum)
gg1
log(sum1)
gg2
gg/1-hh
gg3
gg1/1-hh
e(t)
gg2+gg3
Stop
0,1,2.L-2
hh
m1,m2
2
0,1,2.L-1
Figure 4.4 Flow chart for Renyi Entropy
M. Tech. (CE)
Page 49
2014-15
4.3.2.3 Flow Chart for Havrda-Charvat Entropy

Start
l=256
alpha=0.5
fac=2^(1-alpha)
e(t)=0, k(t)=0
For m1=1:1:t
For t=1:254
For m2=t+1: 1: 255
For m1=t+1:1:255
e(t)=e(t)+(p(m1,m2)âlpha)
For m2=1:1:t
Symbols
Meaning
e(t)
t L 1
( p m1 ,m2 )
m1 0m2 t 1
k(t)
L 1 t
( p m1 ,m2 )
m1 t 1 m2 o
ae1(t)
e(t)+k(t)
ae2(t)
ae1(t)-1
ae(t)
ae2(t)/fac-1
0,1,2.L-2
m1,m2
0,1,2.L-1
k(t)=k(t)+(p(m1,m2))âlpha)
ae1(t)=e(t)+k(t)
ae2(t)=ae1(t)-1
ae(t)=ae2(t)/fac-1
n=1:1:255
Plot(n,ae)
Stop
Figure 4.5 Flow chart for Havrda-Charvat Entropy
M. Tech. (CE)
Page 50
2014-15
4.3.2.4 Flow Chart for Kapur Entropy

Start
l=256
alpha=0.5, beta=3
fac=2^(1-alpha)
e(t)=0, r(t)=0
For t=1:254
For m2=t+1: 1: 255
For m1=1:1:t
For m1=t+1:1:255
e(t)=e(t)+(p(m1,m2)^(alpha+b
eta-1))
r(t)=r(t)+(p(m1,m2)^ beta)
For m2=1:1:t
k(t)=k(t)+(p(m1,m2)^(alpha+beta
-1))
q(t)=q(t)+(p(m1,m2)^ beta)
ns1(t)=e(t)/r(t)
Symbols
Meaning
t L 1 1
Pm1 ,m2
m101 m2 t 1
t 255
( p m1 ,m2 )
m1 1 m2 t 1
e(t)
r(t)
ns1(t)
L 1 t 1
Pm1 ,m2
m1t 1 m2 1
255 t
( p m ,m )
m t 1 m 1
e(t)/r(t)
ns2(t)
k(t)/q(t)
ns3(t)
ns1(t)+ns2(t)
ns11(t)
ns3(t)-2/fac-1
0,1,2.L-2
ns2(t)=k(t)/q(t)
k(t)
ns3(t)=(ns1(t)+ns2(t))
q(t)
ns11(t)=(ns3(t)-2)/fac-1
ae=ns11
Plot(n,ae)
Stop
m1,m2
0,1,2.L-1
Figure 4.6 Flow chart for Kapur Entropy
M. Tech. (CE)
Page 51
2014-15
4.3.2.5 Flow Chart for Vajda Entropy

Start
l=256,hh=2
alpha=0.5, beta=1
fac=2^(1-alpha)
e(t)=0, r(t)=0
For t=1:254
For m2=t+1: 1: 255
For m1=1:1:t
For m1=t+1:1:255
e(t)=e(t)+(p(m1,m2)âlpha
r(t)=r(t)+(p(m1,m2)
For m2=1:1:t
k(t)=k(t)+(p(m1,m2)âlpha
q(t)=q(t)+p(m1,m2)
ns1(t)=e(t)/r(t)
Symbols
t L 1
Pm1 ,m2
m101 m2 t 1
t
255
( p m1 ,m2 )
m1 1 m2 t 1
e(t)
ns2(t)=k(t)/q(t)
r(t)
ns3(t)=(ns1(t)+ns2(t))
L 1

m1t 1
k(t)
ns11(t)=(ns3(t)-2)/fac-1
Plot(n,ae)
m2 1
m1 , m2
q(t)
255 t
( p m1 ,m2 )
m1 t 1 m2 1
ns1(t)
e(t)/r(t)
ns2(t)
k(t)/q(t)
ns3(t)
ns1(t)+ns2(t)
ns11(t)
ns3(t)-2/fac-1
0,1,2.L-2
m1,m2
0,1,2.L-1
ae=ns11
n=1:1:254
Meaning
Stop
Figure 4.6 Flow chart for Vajda Entropy

M. Tech. (CE)
Page 52
2014-15
4.4 Details of Software:

The various authors in the literature review process has experimented his various
approaches over the platforms like Ubuntu, Windows using MATLAB, WEKA. Below
are the details of the software and hardware that are used during the experimentation
process.
4.4.1 Software specification:
The experimentation was carried out on the MATLAB software. MATLAB (matrix
laboratory) is an interactive software package which was developed to perform
numerical calculations on vectors and matrices. It can do quite sophisticated graphics in
two and three dimensions. It contains a high- level programming language (a baby C)
which makes it quite easy to code complicated algorithms involving vectors and
matrices. It can numerically solve nonlinear initial-value ordinary differential equations.
It contains a wide variety of toolboxes which allow it to use in a wide range of
applications.
4.5 Performance Parameters Used

Computations were performed in MATLAB for all texture samples. Below are
described all the features used in the experiment and the meaning of each one in the
actual texture analysis case is explained.
4.5.1 Contrast
Contrast is a local gray level variation in the gray level co-occurrence matrix. It can be
thought of as a linear dependency of grey levels of neighboring pixels.
Contrasti ji j2p( i, j)
i and j are horizontal and vertical cell coordinates and p is the cell value. If the
neighboring pixels are very similar in their gray level values then contrast in the image
is very low. In case of texture, the gray level variations show the variation of texture
itself. High contrast values are expected for heavy textures and low for smooth, soft
textures.
4.5.2 Homogeneity
Homogeneity measures the uniformity of the non-zero entries in the GLCM. It weights
values by the inverse of contrast weight
M. Tech. (CE)
Page 53
Homogeneity =
1 (i j)
2014-15
p(i, j )
i, j
The GLCM homogeneity of any texture is high if GLCM concentrates along the
diagonal, meaning that there are a lot of pixels with the same or very similar grey level
value. The larger the changes in grey values, the lower the GLCM homogeneity making
higher the GLCM contrast. If the image has little variation then homogeneity is high
and if there is no variation then homogeneity is equal to 1. Therefore, high homogeneity
refers to textures that contain ideal repetitive structures, while low homogeneity refers
to big variation in both, texture elements and their spatial arrangements. An
inhomogeneous texture refers to an image that has almost no repetition of texture
elements and spatial similarity in it is absent.
4.5.3 Dissimilarity
Dissimilarity is a measure that defines the variation of grey level pairs in an image. It is
the closest to Contrast with a difference in the weight Contrast unlike Dissimilarity
grows quadratically
Dissimilarity=
i j p(i, j )
i, j
It is expected that these two measures behave in the same way for the same texture
because they calculate the same parameter with different weights. Contrast will always
give slightly higher values than Dissimilarity.
4.5.4 Entropy
Entropy in any system represents disorder, where in the case of texture analysis is a
measure of its spatial disorder
Entropy p(i, j ) log( p(i, j ))

i, j
This feature can be useful to tell us if the entropy is bigger for a heavy texture or for the
smooth textures giving us information about which type of texture can be considered
statistically.
In this chapter the details of system block diagram of the target work, details of
Input/output parameters that was used during experimentations, proposed algorithm and
details of softwares which have been selected to carry out experimentation. In next
M. Tech. (CE)
Page 54
2014-15
chapter includes final objective set, description of experiments that was carried out and
results obtained are presented.
M. Tech. (CE)
Page 55
2014-15
Chapter-5 Experimental Results and Analysis

Previous chapter discussed about the design flow of the targeted work and the details of
data used and performance parameter and also that has been used in the dissertation.
This chapter discusses about the experimental result and provides the comparative
analysis of performance of various entropy measures. The volume of the information
provides computation processing to focus on brain tumor detection at early stages. The
chapter also discusses about the Final objectives and the experiment detail and their
result are explored. After than the experiment variation is evaluated and provide the
validity of result and the information about the contemporize works.
5.1 Experimentation Scenarios:
The experiment is carried out using the Shannon and Non-Shannon entropy measures in
order to recognize the existence of tumor in brain
5.1.1 Experiment No. 1
To Perform the Shannon Entropy Measure in order to detect the Tumor.
Table 5.1 Shannon Entropy
Grade 1
MRI Image
I/P Image
Graph
Entropy values
Tumor
Grades
O/P Image
Gray Scale values
Threshold=94
M. Tech. (CE)
Page 56
I/P Image
Grade 2
2014-15
Entropy values
Gray Scale values
O/P Image
I/P Image
Entropy values
Grade 3
Threshold=110
Gray Scale values
O/P Image
Entropy values
I/P Image
Grade 4
Threshold=121.2
Gray Scale values
O/P Image
Threshold=126
M. Tech. (CE)
Page 57
2014-15

To Perform the Vajda Entropy Measure in order to detect the Tumor.
Table 5.2 Vajda Entropy
Grade 1
MRI Image
I/P Image
Graph
Entropy values
Tumor
Grades
Gray Scale values
Threshold=94.7
I/P Image
Grade 2
Entropy values
O/P Image
Gray Scale values
O/P Image
I/P Image
Entropy values
Grade 3
Threshold=102
Gray Scale values
O/P Image
Threshold=123.3
M. Tech. (CE)
Page 58
I/P Image
Grade 4
O/P Image
2014-15
Entropy values
Gray Scale values
Threshold=143
In the Vajda Entropy, it is evaluated that the Tumor is detecting at Stage 3 and Stage4.
The Entropy value of the graph at different stages varies accordingly. While scanning
the Grade1 Tumor MRI Image the entropy value is 94.7. While the Entropy value for
Grade2, Grade3 and Grade4 types Tumor are 102, 123.3 and 143.
M. Tech. (CE)
Page 59
2014-15
5.1.3 Experiment No.3

To Perform the Kapur Entropy Measure in order to detect the Tumor.
Table 5.3 Kapur Entropy
MRI Image
Grade 1
I/P Image
Graph
Entropy values
Tumor
Grades
Gray Scale values
Threshold=98
I/P Image
Grade 2
Entropy values
O/P Image
Gray Scale values
Threshold =103.7
Grade 3
I/P Image
Entropy values
O/P Image
Gray Scale values
Threshold =143.2
O/P Image
M. Tech. (CE)
Page 60
Grade 4
2014-15
I/P Image
Gray Scale values
Threshold =168
O/P Image
In this, it is evaluated that the Tumoris detecting at Stage 3 and Stage4. The Entropy
value of the graph at different stages varies accordingly. While scanning the Grade1
Tumor MRI Image the entropy value is 98. While the Entropy value for Grade2, Grade3
and Grade4 types Tumor are 103.7, 143.2 and 168.
5.1.4 Experiment No.4
To Perform the Renyi Entropy Measure in order to detect the Tumor.
Table 5.4 Renyi Entropy
MRI Image
Grade 1
I/P Image
Graph
Entropy values
Tumor
Grades
Gray Scale values
O/P Image
Threshold=103
M. Tech. (CE)
Page 61
I/P Image
Grade 2
2014-15
Entropy values
Gray Scale values
Threshold =130
Grade 3
I/P Image
Entropy values
O/P Image
Gray Scale values
Threshold =164.2
Grade 4
I/P Image
Entropy values
O/P Image
Gray Scale values
Threshold =191
O/P Image
In this, it is evaluated that the Tumor is detecting at Stage 2, Stage 3 and Stage4. The
Entropy value of the graph at different stages varies accordingly. While scanning the
M. Tech. (CE)
Page 62
2014-15
Grade1 Tumor MRI Image the entropy value is 103. While the Entropy value for
Grade2, Grade3 and Grade4 types Tumor are 130, 164.2 and 191.
To Perform the Havrda Charvat Entropy Measure in order to detect the Tumor.
Table 5.5 Havrda-Charvat Entropy
MRI Image
Grade 1
I/P Image
Graph
Entropy values
Tumor
Grades
Gray Scale values
O/P Image
I/P Image
Grade 2
O/P Image
Entropy values
Threshold =115
Gray Scale values
Threshold =127.9
M. Tech. (CE)
Page 63
Grade 3
I/P Image
2014-15
Entropy values
O/P Image
Gray Scale values
Grade 4
I/P Image
Entropy values
Threshold =153.8
Gray Scale values
O/P Image
Threshold =164
In this, it is evaluated that the Tumor is detecting at Stage1. So, all other Tumor at
different stages will definitely get detected. The Entropy value of the graph at different
stages varies accordingly. While scanning the Grade1 Tumor MRI Image the entropy
value is 115. While the Entropy value for Grade2, Grade3 and Grade4 types Tumor are
127.9, 153.8 and 164.
5.2 Result Analysis

5.2.1 Result analysis through graph
In this section the results of the various entropy measures are compared using its graph
and the entropy values which are different for same image in different entropies.
M. Tech. (CE)
Page 64
2014-15
Table 5.6 Result analysis using different entropies
Image
Havrda
Charvat
Graph
Threshold
Values
168
Entropy values
Entropy
Renyi
Entropy values
Gray Scale values
143
Kapur
Entropy values
Gray Scale values
123.2
Vajda
Entropy values
Gray Scale values
98.4
Gray Scale values
M. Tech. (CE)
Page 65
Entropy values
Shannon
2014-15
94
Gray Scale values
In this table it can be seen that how the entropy values are changing for the same image
in different entropy. It is also depending on the fact that higher the entropy value, fast
the tumor detection.
5.2.2 Result analysis through MRI Images
In this section the discussion on the output of the various entropies has been done. The
variation of results under different entropies can be seen in the table. The corresponding
entropy values of this table are displayed in next table.
Table 5.7 Output Images
Case
Imag
e No.
MRI
Images
(I/P
Images)
Havrda
Charvat
(O/P
Images)
Renyi
(O/P
Images)
Kapur
(O/P
Images)
Vajda
(O/P
Images)
Shannon
(O/P
Images)
1a
Grade
1
1b
1c
M. Tech. (CE)
Page 66
2014-15
2a
Grade
2
2b
2c
3a
Grade
3
3b
3c
4a
Grade
4
4b
4c
The table shows the improvement of information obtained through images and it was
analyze that Havrda Charvat is producing the best result while detecting the Tumors at
M. Tech. (CE)
Page 67
2014-15
Stage1. Other than Havrda Entropy the Renyi Entropy is detecting the Tumor at stage2
of Tumor. The rest of the entropies are detecting the Tumor at its last stage.
5.2.3 Result Analysis using Texture Features
Tabular formation of the texture features of the different MRI Images that is the
Contrast, Homogeneity, Dissimilarity, and Entropy are shown below in the table for the
corresponding entropies.
Table5.8 Texture Feature of different cases for Shannon Entropy
Case
Grade1
Grade2
Grade3
Grade4
Image
No.
1a
1b
1c
2a
2b
2c
3a
3b
3c
4a
4b
4c
Contrast
179.5465
220.2310
224.3410
194.0014
229.7970
231.5867
225.5463
239.4755
245.0573
256.2873
258.7493
224.6723
Dissimilarity
1.3902
1.6500
1.8902
2.0670
2.1591
2.0185
2.6501
3.2006
3.4367
4.2463
4.3384
4.7372
Homogeneity
0.0227
0.1845
0.1227
0.2459
0.2145
0.2091
0.3145
0.3624
0.4020
0.4144
0.4270
0.4514
Entropy
0.4622
0.3422
0.4012
0.2474
0.1422
0.0538
0.1322
0.2962
0.3782
1.3529
1.2217
1.3147
Table 5.9 Texture Feature of different cases for Renyi Entropy
Case
Grade1
Grade2
Grade3
Grade4
Image
No.
1a
1b
1c
2a
2b
2c
3a
3b
3c
4a
4b
4c
Contrast
212.8751
211.7970
229.2630
230.5965
250. 5965
249.5867
263.7971
269.4755
270.0573
278.2873
235.7493
271.6723
Dissimilarity
3.8402
3.8500
3.8902
4.0670
4.3501
4.3185
4.4500
5.2516
5.2532
5.3704
6.5419
7.0071
M. Tech. (CE)
Homogeneity
0.3107
0.3145
0.4227
0.5059
0.5105
0.5612
0.6215
0.6301
0.6621
0.7044
0.7110
0.7164
Entropy
0.4622
0.3422
0.4622
1.3474
1.3422
1.4538
1.1422
2.4262
2.7782
1.8529
3.0217
3.3647
Page 68
2014-15
Table 5.10 Texture Feature of different cases for Kapur Entropy
Case
Grade1
Grade2
Grade3
Grade4
Image
No.
1a
1b
1c
2a
2b
2c
3a
3b
3c
4a
4b
4c
Contrast
209.1426
211.7970
219.2635
221.7281
241.7971
271.5867
251.2718
289.4755
305.0573
262.1728
305.7492
301.6720
Dissimilarity
2.8302
2.8500
3.1546
3.1795
3.1928
3.7585
4.0180
4.1696
4.4364
5.0956
5.5465
6.5464
Homogeneity
0.2027
0.3145
0.3227
0.3459
0.4125
0.4791
0.4911
0.5218
0.5620
0.6294
0.6570
0.6951
Entropy
0.2122
0.3022
0.4152
0.3004
0.3122
0.4868
1.1092
1.2022
1.4712
2.3519
2.4177
3.3917
Table 5.11 Texture Feature of different cases for Vajda Entropy
Case
Grade1
Grade2
Grade3
Grade4
Image
No.
1a
1b
1c
2a
2b
2c
3a
3b
3c
4a
4b
4c
Contrast
204.1627
242.7970
249.5630
208.6473
257.6170
256.6167
250.2637
289.4395
265.9173
259.3738
245.7493
271.6703
Dissimilarity
1.9142
2.3500
2.4912
2.8670
3.4511
3.6185
4.1500
4.2576
4.3532
5.0704
5.5119
6.0071
Homogeneity
0.1264
0.1752
0.2227
0.2459
0.3145
0.3791
0.5019
0.5215
0.5620
0.5944
0.6370
0.6514
Entropy
0.3622
0.3422
0.3622
0.4474
0.4422
0.5485
1.5474
1.6035
2.6192
1.6237
2.7586
2.6474
Table 5.12 Texture Feature of different cases for Havrda-Charvat Entropy
Case
Grade1
Grade2
Image
No.
1a
1b
1c
2a
2b
2c
3a
Contrast
259.2630
259.7970
260.2630
261.5965
265.7970
290.5867
297.7970
Dissimilarity
4.7182
4.7384
4.8002
5.0670
5.6373
5.7185
5.8500
M. Tech. (CE)
Homogeneity
0.5117
0.5145
0.5427
0.5741
0.6145
0.6593
0.7182
Entropy
1.2822
2.2782
1.2432
2.4721
2.3401
2.3738
1.2922
Page 69
Grade3
Grade4
3b
3c
4a
4b
4c
309.4755
335.0573
326.2873
365.7493
361.6723
6.2576
6.9532
7.0704
7.5419
7.0071
0.7196
0.7670
0.7944
0.8250
0.8514
2014-15
1.3162
3.1782
2.8104
2.0217
3.2647
5.3.5 Performance Analysis Bar Chart

The bar chart represents the performance evaluation of the different entropies. After
performing the experimentation work it was observed that in detecting the tumor at its
earlier stage the Havrda Charvat produces the best result in terms of preservation of
color and to get an accurate picture. After then the Renyi Entropy gives its best while
detecting the Tumor at stage2. Other than these two entropies the other entropies detect
the tumors at its last stage that is when the tumor becomes malignant.
350
300
Threshold Values
250
Kapur Entropy
200
Renyi Entropy
150
Vajda entropy
Havrda Charvat Entropy
100
Shannon Entropy
50
0
Grade1 Tumor
Grade2 Tumor
Grade3 Tumor
Grade4 Tumor
Tumor Grades
Fig. 5.1 Performance Analysis Bar Chart
The bar chart that is represented in the fig. 5.1 is on the basis of following data. The
table shows the different entropy values for the different data.
M. Tech. (CE)
Page 70
2014-15
Table 5.13 Data Analysis of different Entropy Function
Stages
Havrda
Renyi
Kapur
Vajda
Shannon
Threshold
= 276.8
Threshold
=142.2
Threshold
= 136.98
Threshold =
122.7
Threshold =
110
{Detectin
g Tumor}
{No
Tumor
Detection
}
{No Tumor {No Tumor

Detection } Detection }
{No Tumor
Detection }
Threshold
= 284.87
Threshold
= 273.62
Threshold
= 150.51
Threshold =
148.87
Threshold =
132.51
{Detectin
g Tumor}
{Detectin
g Tumor}
{No Tumor {No Tumor

Detection } Detection }
{No Tumor
Detection }
Entropy=
298.76
Entropyee Entropyee= Entropyee=

= 280.76
271.76
247.76
Entropyee=
144.76
{Detectin
g Tumor}
{Detectin
g Tumor}
{No Tumor
Detection }
Stage1
Stage 2
{Detecting
Tumor}
{ Detecting
Tumor }
Stage 3
Entropyee Entropyee Entropyee= Entropyee=264.
= 315.76
= 297.76
275.76
76
Entropyee=25
5.76
{Detectin
g Tumor}
{Detecting
Tumor}
{Detectin
g Tumor}
{Detecting
Tumor}
{Detecting
Tumor}
Stage 4
M. Tech. (CE)
Page 71
2014-15
5.3.6 Graphical Representation of the Performance Parameters

The representation of the parameters graphically indicates visually the ranges that were
tabularly obtained and makes the study more clear.
1. Contrast: The contrast of the abnormal image is having the minimum value i.e.
250.5965 and this value are not within the range of normal image
350
300
Contrast
250
200
Grade1
150
Grade2
100
Grade3
Grade4
50
0
Shannon
Renyi
kapur
Havrda
Charvat
Vajda
Entropy Functions
Fig. 5.2 Representation of Contrast Values
The bar chart that is represented in the fig. 5.2 is on the basis on the following data. The
table shows the different contrast values for the different data.
Table 5.14 Analysis of Contrast Values
Entropy
Shannon
Renyi
Havrda-Charvat
Kapur
Vajda
Grade1
179.5465
212.8751
259.263
209.1426
204.1627
Contrast Values
Grade2
Grade3
194.0014
225.5463
250.5965
263.7971
261.5965
297.797
221.7281
251.2718
208.6473
250.2637
Grade4
256.2873
278.2873
326.2873
262.1728
259.3738
2. Homogeneity: The homogeneity of the abnormal image is having the minimum

value i.e. 0.5 and this value are not within the range of normal image.
M. Tech. (CE)
Page 72
2014-15
1
0.9
0.8
Homogeneity
0.7
0.6
Grade1
0.5
Grade2
0.4
Grade3
0.3
Grade4
0.2
0.1
0
Shannon
Renyi
Kapur
Havrda-Charvat
Vajda
Entropy Functions
Fig. 5.3 Representation of homogeneity values
The bar chart that is represented in the fig. 5.1 is on the basis of following data. The
table shows the different homogeneity values for the different data.
Table 5.15 Analysis of Homogeneity Values
Homogeneity
Entropy Measures
Grade1
Grade2
Grade3
Grade4
Shannon
Renyi
Kapur
Havrda-Charvat
Vajda
0.0227
0.3107
0.2027
0.5117
0.1264
0.2459
0.5059
0.3459
0.5741
0.2459
0.3145
0.6215
0.4911
0.7196
0.5019
0.4144
0.7044
0.6294
0.7944
0.5944
Above the representation of the parameters done graphically that indicates visually the
ranges that were tabularly obtained and makes the study more clear. This way the
difference between the different Entropy Functions was evaluated and the result was
observed that Havrda-Charvat Entropy Function produced more accurate result than any
other entropy functions.
5.4 Future Work
The research work that can be carried out in future is introduced in this section.
To incorporate atlas based prior information may prove useful for segmentation
of more subtle and complex tumors.
M. Tech. (CE)
Page 73
2014-15
To employ the automatic symmetry axis detection and studying the influence of
possible non-symmetries caused by the patient head position during MR
scanning.
In this chapter various experimental results for all entropies for tumor detection have
been done. From all the work it can be concluded that Havrda-Charvat entropy produces
the best result in terms of detection of Tumor from MRI images in its early stage. It also
Provide the best color preservation in the output images other than the rest of the
Entropy Measures.
M. Tech. (CE)
Page 74
2014-15
CHAPTER 6 - CONCLUSION
In the field of medical image processing, detection of brain tumor from magnetic
resonance image (MRI) brain scan has become one of the most active Researches. For
accurate diagnosis of brain tumor, proper segmentation method is required to be used
for MRI Images to carry out an improved diagnosis and treatment. Currently,
information is provided by many images from various slices required for accurate
diagnosis, planning and treatment purpose. The volume of the available information
requires computation processing to inform the decision-making. Now-e-day, speed of
computation is no longer an issue for researchers. Therefore, the focus is directed
toward improvement of information from images obtained through the slice orientation
and perfecting the process of segmentation to get an accurate picture of the brain tumor.
Through analysis of the literature it was found that automation of brain tumor detection
and segmentation from brain MRI Images is one of the most active research areas and
enormous research has been done in this area for the last many years.
The thesis attempts at developing an approach so that the early detection and
classification of Brain Tumor become possible. First of all the co-occurrence matrix of
the image is to be segmented and computed for each color channel. For the
segmentation, the proposed method combines the information of edge and of region to
obtain the good result. The probability distribution is then calculated from its cooccurrence matrix. The entropy function of a entropy is defined in entropy definitions,
are then calculated for each image to be segmented using the probability distribution.
The numbers of minima points are determined from the entropy function and the
entropy values were calculated. The improvement of information obtained from images
and perfecting the process of segmentation to get an accurate picture of the brain tumor
must be given importance. The target area is segmented and the evaluation of the nature
of the Tumor is diagnosed and state of the tumor is monitored.
The proposed tumor detection and localization system was found to be able to
accurately detect and localize brain tumor in magnetic resonance imaging. Since the
number of images of healthy brain is not much higher than the number of images of
effected brain, the overall accuracy of the system can also be expressed. We were able
to extract the useful information for detecting the brain tumors.
M. Tech. (CE)
Page 75
2014-15
There are also many other methods used for tumor extraction but they usually rely on
machine learning algorithms such as SVM. The big advantage of this method that was
observed is that the simulation results for different entropy measures provide better
results in the sense of preservation of color in different segments.
M. Tech. (CE)
Page 76
2014-15
REFERENCES
[1] Amsaveni, V.; Singh, N.A.; Dheeba, J., "Computer aided detection of tumor in MRI
brain images using cascaded correlation neural network," Sustainable Energy and
Intelligent Systems (SEISCON 2013), IET Chennai Fourth International Conference
on , pp.527,532, 12-14 Dec. 2013 doi: 10.1049/ic.2013.0365
[2] Abbasi, S.; Tajeri Pour, F., "A hybrid approach for detection of brain tumor in MRI
images,"Biomedical Engineering (ICBME), 2014 21th Iranian Conference on ,
pp.269,274, 26-28 Nov. 2014 doi: 10.1109/ICBME.2014.7043934
[3] Abdullah, A.A.; Chize, B.S.; Nishio, Y., "Implementation of an improved cellular
neural network algorithm for brain tumor detection," Biomedical Engineering (ICoBE),
2012 International Conference on , vol., no., pp.611,615, 27-28 Feb. 2012
doi: 10.1109/ICoBE.2012.6178990
[4] Atkins,M.S.; Mackiewich, B.T., "Fully Automatic Segmentation of the Brain in
MRI," Engineering in Medicine and Biology Society. IEEE-EMBS 1998. International
Transaction on , pp.5104,5107, 17-18 Jan. 2015doi: 10.1109/IEMBS.1998.161562
[5] Chandra, S.; Bhat, R.; Singh, H., "A PSO based method for detection of brain
tumors from MRI,"Nature& Biologically Inspired Computing, 2009. NaBIC 2009.
World Congress on , pp.666,671, 9-11 Dec. 2009 doi: 10.1109/NABIC.2009.5393455
[6] CC Leung; W.F. Chen ;F.H.Y Chon., "Brain Tumor Boundary Detection In MRI
Image with Generalised Fuzzy Operator,"NaBIC 2009. World Congress on ,
pp.666,671, 9-11 Dec. 2003 doi: 10.1109/NABIC.2003.7803-7750
[7] Dvorak, P.; Kropatsch, W.; Bartusek, K., "Automatic detection of brain tumors in
MRI Images,"Telecommunications and Signal Processing (TSP), 2013 36th
International
Conference
on ,
pp.577,580,
2-4
July
2013
doi:
10.1109/TSP.2013.6614000
[8] Dhage, P.; Phegade, M.R.; Shah, S.K., "Watershed segmentation brain tumor
detection," Pervasive Computing (ICPC), 2015 International Conference on , pp.1,5, 810 Jan. 2015 doi: 10.1109/PERVASIVE.2015.7086967
[9] GladisPushpaRathi, V.P.; Palani, S., "Detection and characterization of brain tumor
using segmentation based on HSOM, wavelet packet feature spaces and
ANN," Electronics Computer Technology (ICECT), 2011 3rd International Conference
on , vol.6, pp.274,277, 8-10 April 2011 doi: 10.1109/ICECTECH.2011.5942097
[10] Ghanavati, S.; Junning Li; Ting Liu; Babyn, P.S.; Doda, W.; Lampropoulos, G.,
"Automatic brain tumor detection in Magnetic Resonance Images," Biomedical Imaging
(ISBI), 2012 9th IEEE International Symposium on , pp.574,577, 2-5 May 2012 doi:
10.1109/ISBI.2012.6235613
M. Tech. (CE)
Page 77
2014-15
[11] Islam, A.; Reza, S.M.S.; Iftekharuddin, K.M., "Multifractal Texture Estimation for
Detection and Segmentation of Brain Tumors," Biomedical Engineering, IEEE
Transactions
on ,
vol.60,
no.11,
pp.3204,3215,
Nov.
2013
doi:
10.1109/TBME.2013.2271383
[12] Jason J Corso, A.; Reza, S.M.S.; K.M., "Efficient Multilevel Brain Tumor
Segmentation with Integrated Bayesian Model Classification," Biomedical Engineering,
IEEE Transactions on , vol.27, no.5, pp. 0278-0062/$25.00, Nov. 2008 doi:
10.1109/TBME.2013.2271383
[13] Kharrat, A.; Benamrane, N.; Ben Messaoud, M.; Abid, M., "Detection of brain
tumor in medical images," Signals, Circuits and Systems (SCS), 2009 3rd International
Conference on , pp.1,6, 6-8 Nov. 2009 doi: 10.1109/ICSCS.2009.5412577
[14] K.K. Pawan; CharuMadhawan; D.C. Chaturvedi"Automatic detection of brain
tumors in MRI Images,"Telecommunications and Signal Processing (TSP), 2013 36th
International
Conference
on ,
pp.577,580,
2-4
July
2013
doi:
10.1109/TSP.2013.6614000
[15] Logeswari, T.; Karnan, M., "An Enhanced Implementation of Brain Tumor
Detection Using Segmentation Based on Soft Computing," Signal Acquisition and
Processing, 2010. ICSAP '10. International Conference on , pp.243,247, 9-10 Feb. 2010
doi: 10.1109/ICSAP.2010.55
[16] Leung, C.C.; Chen, W.F.; Kwok, P.C.K.; Chan, F.H.Y., "Brain tumor boundary
detection in MRI Image with generalized fuzzy operator," Image Processing, 2003.
ICIP 2003.Proceedings. 2003 International Conference on , vol.2, pp.II,1057-60 vol.3,
14-17 Sept. 2003 doi: 10.1109/ICIP.2003.1246867
[17] Murugesan, M.; Sukanesh, R., "Automated Detection of Brain Tumor in EEG
Signals Using Artificial Neural Networks," Advances in Computing, Control, &
Telecommunication Technologies, 2009. ACT '09. International Conference on ,
pp.284,288, 28-29 Dec. 2009 doi: 10.1109/ACT.2009.77
[18] Parveen; Singh, A., "Detection of brain tumor in MRI images, using combination
of fuzzy c-means and SVM," Signal Processing and Integrated Networks (SPIN), 2015
2nd International Conference on , pp.98,102, 19-20 Feb. 2015 doi:
10.1109/SPIN.2015.7095308
[19] Phooi Yee Lau; Voon, F.C.T.; Ozawa, S., "The detection and visualization of brain
tumors on T2-weighted MRI images using multiparameter feature blocks," Engineering
in Medicine and Biology Society, 2005. IEEE-EMBS 2005. 27th Annual International
Conference of the , pp.5104,5107, 17-18 Jan. 2015doi: 10.1109/IEMBS.2005.161562
[20] PadmakantDhage; M.R. Phegde; S.K. Shah, "Watershed segmentation Brain
Tumor Detection," Signal Processing and Integrated Networks (SPIN), 2015 2nd
M. Tech. (CE)
Page 78
2014-15
International Conference on , pp.98,102, 19-20 Feb. 2015 doi: 10.1109/SPIN.2015.6272
[21] Saha, S.; Das, S.K.; Kar, A., "A new segmentation technique for brain and head
from high resolution MRI Image using unique histogram features," Image Processing
Theory Tools and Applications (IPTA), 2010 2nd International Conference on ,
pp.288,293, 7-10 July 2010 doi: 10.1109/IPTA.2010.5586722
[22] Sasikala, M.; Kumaravel, N., "Comparison of Feature Selection Techniques for
Detection of Malignant Tumor in Brain Images," INDICON, 2005 Annual IEEE ,
pp.212,215, 11-13 Dec. 2005 doi: 10.1109/INDCON.2005.1590157
[23] Satish Chandra, Rajesh Bhatt, "A PSO Based Method for detection of Brain
Tumor," Electronics Computer Technology (ICECT), 2009 3rd International
Conference
on ,
vol.6,
pp.274,277,
8-10
April
2009
doi:
10.1109/ICECTECH.2009.42445612
[24] SaharGhanvati, Junning Li, A., "Automatic Brain Tumor Detection in Magnetic
Resonance Images," Image Processing Theory Tools and Applications (IPTA), 2012
2nd International Conference on ,
pp.288,293, 7-10 July 2012doi:
10.1109/IPTA.2012.1451858
[25] SolmazAbbasi; FarshadTajeriPaur, "A Hybrid Approach for Detection of Brain
Tumors in MRI Images," Signals, Circuits and Systems (SCS), 2009 3rd International
Conference on , pp.1,6, 6-8 Nov. 2014doi: 10.1109/ICSCS.2014.78-1-4799-7418
[26] Somwanshi, D.K.; Goswami, A., "Detection of Abnormalities in MRI Images using
Texture Analysis," Journal of International Academy of Physical Sciences, Special Issue
2, pp. 339-346, Vol. 15,14-16 June 2011.
[27] Vijay, J.; Subhashini, J., "An efficient brain tumor detection methodology using Kmeans clustering algoriftnn," Communications and Signal Processing (ICCSP), 2013
International
Conference
on ,
pp.653,657,
3-5
April
2013
doi:
10.1109/iccsp.2013.6577136
[28] Wang, T; Karnan, M., "An Enhanced Implementation of Brain Tumor Detection
Using Segmentation Based on Soft Computing," Signal Acquisition and Processing,
2010. ICSAP '10. International Conference on , pp.243,247, 9-10 Feb. 2010 doi:
10.1109/ICSAP.2010.55
[29] Wang Yang; Ma Siliang, "Automatic detection and segmentation of brain tumor
using fuzzy classification and deformable models," Biomedical Engineering and
Informatics (BMEI), 2011 4th International Conference on , vol.3, pp.1680,1683, 15-17
Oct. 2011 doi: 10.1109/BMEI.2011.6098610
[30] XiaoliXu; Yang Hu, An Efficient Brain Tumor Detection Methodology using KMeans Image Processing Theory Tools and Applications (IPTA), 2012 2nd
M. Tech. (CE)
Page 79
International
Conference
10.1109/IPTA.2012.1451858
on ,
pp.288,293,
M. Tech. (CE)
7-10
July
2014-15
2012
doi:
Page 80

Marketing

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Marketing

Uploaded by

Copyright:

Available Formats

An Efficient Brain Tumor Extraction from MRI Images using Entropy Measures

Fig. 1.1 The Major Subdivision Of Human Brain [19]

Following are the main parts of the human brain:

Poornima University, Jaipur

1.2 Brain Tumors

1.3 MRI Images

Fig. 1.3Patients Head Examined in a Clinical Diagnosis [21]

Fig. 1.3.1 (a) Axial Plane

(b) Sagittal Plane

(c) Coronal Plane [21]

Poornima University, Jaipur

reproducibility comparable to manual segmentation. In addition, it also reduces the time

1.4 The Image Segmentation

1.5 Application of Image Segmentation

Measure tissue volumes

Diagnosis, study of anatomical structure

Virtual surgery simulation

1.5.4 Object Detection: Object detection is a computer technology related to computer

Brake light detection

Locate objects in satellite images (roads, forest, crops, etc.)

Poornima University, Jaipur

1.6 Introduction to Brain Tumor Segmentation

1.7 Difficulties in Segmentation of Brain MRI

Poornima University, Jaipur

1.9 Purpose of Thresholding

1.10 Thresholding Algorithm

1.11 Thresholding Methods

Poornima University, Jaipur

Chapter2: Literature Review

2.1 Review Process Adopted

Stage 2: Get the Details.

Fig. 2.1 Review Process Stage Analysis

The details of various stages followed are discussed below.

process on more than 70 papers.

Stage 3+:Synthesize the detail

2.2 Categorical Reviews in An efficient Brain Tumor extraction from

S.NO. Name of Issue

Total Paper Number of Paper Reviewed

detection from MRI

However, in order to obtain better results from the proposed

Poornima University, Jaipur

Poornima University, Jaipur

Poornima University, Jaipur

Poornima University, Jaipur

Poornima University, Jaipur

Poornima University, Jaipur

The feed forward back propagation neural networks provide computationally

Feature selection algorithms were categorized into Exponential, Randomized

Poornima University, Jaipur

The segmentation and the localization of suspicious regions were performed by

A mathematical morphology was adopted to increase the contrast in MRI

The Hybrid algorithm used Contour-based and Region based Methods to

The performance of clustering algorithm for image segmentation was highly

2.3 Various solution approaches in the issue Brain Tumor Extraction

Poornima University, Jaipur

Table 2.3: Categorical Review of Research Paper

[C.C. Leung, et-al,

[Ahmed Kharrat, etal, 2009]

[M. Stella Atkins,etal,1998]

[SaharGhanavati, etal, 2012]

[Wang Yang, etal,2011]

Poornima University, Jaipur

Comparision of Manual and Automatic Brain Segmentation

The extraction of the tumor was accomplished