Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Thesis

on
DETECTION AND CLASSIFICATION OF BRAIN TUMOR

USING SUPPORT VECTOR MACHINE BASED GUI

Submitted in partial fulfilment for the award of degree of

MASTER OF TECHNOLOGY

In

Electronics and Communication Engineering

(Electronic Circuits and Systems)

Submitted By:
Farheen Wahab
(Roll no.: 1701311002)

Under the Guidance of


Dr Imran Ullah Khan
(Jr. Associate Professor)

Department of Electronics & Communication Engineering


INTEGRAL UNIVERSITY, LUCKNOW, U.P. - 226026
May 2019
CERTIFICATE

This is to certify that Farheen Wahab has carried out the research work

presented in the thesis entitled “Detection and Classification of Brain Tumor

using Support Vector Machine based GUI” for the award of Master of

Technology (M.Tech.) in Electronic Circuits and Systems from Department of

Electronics and Communication Engineering, Integral University, Lucknow

under my supervision. To the best of my knowledge, the contents of this thesis have

not been submitted to any other institute or university for the award of any degree.

Signature of Supervisor

Name: Dr. Imran Ullah Khan

Designation: Jr. Associate Professor

Department of ECE

Integral University

Lucknow

Date -

ii
ABSTRACT

Image processing has recently gained a lot of importance in the field of


medical sciences, medical imaging being the main area of research. One of its
major applications is in the detection and diagnosis of diseases. In this work,
the detection of brain tumor is taken under consideration. The vital role in the
entire procedure is played by image segmentation. It has become one of the
challenging tasks in the imaging area, mainly because it includes the
extraction of the desired regions from the images. It plays a significant role
particularly in the medical field where the removal of tumor tissues decides
the extent of success of the diagnosis. Usually, medical experts, for instance, a
surgeon, carries out this task but most of the times it becomes difficult due to
the miscellany nature of the tumor tissues. It also varies from patient to
patient and sometimes depends upon the resemblance between the normal
tissue and a tumor tissue. Hence, the automation of this task of medical image
segmentation is of paramount importance and it remains challenging. This
area has drawn the attention of researchers in the last so many years.

In this work main area of concern will be brain image segmentation.


The main focus is on the detection and extraction of tumor regions, in
particular, from the Magnetic Resonance (MR) brain images. MRI has
become a primary choice of doctors to obtain good quality medical imaging.
The whole problem is dealt as a classification problem to distinguish between
the normal and abnormal brain scans. The procedure is divided into three
major steps, i.e. feature extraction, feature reduction and classification. To
distinguish between the normal and abnormal tissues in the scan, some

iii
features, for instance, first and second order features are calculated from the
images. These features mainly include texture features such as mean, entropy,
correlation, contrast and many others. Discrete wavelet transform is used to
denoise the images and extract the desired features from them. There are
plenty of features which can be obtained from the images, but not all of them
are useful, so there is a need to reduce the features to the desired level.
Principle component analysis (PCA) is used for feature reduction. The main
advantage of using PCA is that it regroups the features according to their
contribution in retaining the variance of the image. All the features are then
checked and the ones making the least contribution are rejected from the list.
In this way, no important feature is lost. The classification problem is treated
by using Support Vector Machines (SVM) which is a well-accepted method
nowadays to handle the similar problems.

Lastly, an interface has been developed for the same algorithm to easily
access the work. The interface requires the user to enter the input image and
the segmented image for the same can be obtained by clicking on the required
push button. All the features will be displayed automatically after the
segmented image is obtained. The entire work has been implemented using
MATLAB R2013a. The experimental work is shown in chapter 5. The
classification analysis shows that the highest accuracy was achieved with the
linear kernel, i.e. about 86.5 %. The RBF, polygonal and quadratic kernels
achieved 78.5 %, 77.5 % and 77.5 % respectively. The time analysis shows
that the entire classification process took about 0.0428 seconds to complete.
This shows that the experiment can be incorporated in real time situations.

iv
ACKNOWLEDGEMENT

“He is the source of all light and His light is diffused throughout the universe.
Allah has given the man intelligent speech, power of expression, and capacity
to understand clearly the relations of things.”

First and foremost I would thank the Almighty whose infinite grace
make all things possible. I am grateful to my parents who have inculcated in
me good values and culture and because of their immense contribution, this
work is dedicated to them as a token of love. We are living in the age of
electronics and this field has witnessed radical changes over the last decades.
We have observed vast application of electronics in almost every walk of life
especially in the field of medical science. This fascinated me to do something
in this field as a student of electronics science.

In my project work great many people have contributed time and


efforts in bringing the project to a conclusion. I feel great pleasure in
expressing my sense of gratitude and sincere thanks to my project guide Dr
Imran Ullah Khan, Department of Electronics and Communication
Engineering for his consistent and careful guidance throughout this project
work. I feel proud to get the opportunity to work under him which has
enlightened my thoughts and enriched my experience. I am grateful to Dr
Syed Hasan Saeed, Professor and Head, Department of Electronics and
Communication Engineering for providing necessary facilities in the
department. I am also thankful to Dr Shailendra Kumar for his constant
support throughout this semester. I shall be failing in my duty if not

v
mentioned the entire teaching fraternity of Department of Electronics and
Communication Engineering for their untiring efforts in educating me.

I owe a debt of gratitude to library staff of the university, my


classmates and friends who helped me directly or indirectly during this
project work.

FARHEEN WAHAB

M.Tech 2nd Year

Roll No 1701311002

Electronics Circuits and Systems

Department of ECE

Integral University

Lucknow

vi
Table of Contents

S No. Title Page No.


Certificate ii
Abstract iii-iv
Acknowledgment v-vi
List of Tables ix
List of Figures x – xi
List of Abbreviations xii-xiii

1. Introduction 2
2. Literature Review 5
2.1 Highlights of the work 9
3. Image Segmentation Basics 12
3.1 Some Imaging Concepts 13
3.2 Images of Brain and their Treatment 15
3.2.1 Magnetic Resonance Imaging 16
3.2.2 Diagnosis of Brain Tumor 17
3.2.3 Diversity of Brain Tumor 21
3.3 Image Segmentation 21
3.4 Conclusion 26
4. Support Vector Machine 28
4.1 Notations 28
4.2 SVM used for Binary Classification 28
4.2.1 Linear Classification with linearly separable
29
problems
4.2.2 Linear Classification with non-separable 35

vii
problems
4.2.3 Non-Linear Classification with non-separable
36
problems
4.3 Multi Class SVM 39
4.4 Conclusion 41
5. Results 43
5.1 Flowchart 44
5.2 Noise Reduction 45
5.3 Feature extraction and reduction 47
5.3.1 First order features 50
5.3.2 Second order features 51
5.4 K-Fold Stratified Cross Validation 54
5.5 Experiments and Discussion 55
5.5.1 Results for Benign Tumor 57
5.5.2 Results for Malignant Tumor 63
5.5.3 Classification Analysis 69
5.5.4 Time evaluation 69
Conclusion And Future Scope 71
References 73
Publications 77
APPENDIX I 78

viii
List of Tables

Table No. Title Page No.


3.1 Tumor Classification with grades 22
4.1 Kernel functions 38
5.1 Division of training and testing images 56
5.2 First order features for benign tumor 62
5.3 Second order features for benign tumor 63
5.4 First order features for malignant tumor 68
5.5 Second order features for malignant tumor 69
5.6 Classification analysis 69

ix
List of Figures

Figure No. Title Page No.


2.1 Flowchart of the proposed method 10
3.1 Grey level image with its matrix representation 14
3.2 Window of a grey level image 14
3.3 Histogram of a grey level image 15
3.4 T1 and T2 weighted images 17
3.5 Before and after of T1 weighted images with
injection of contrast agent 17
3.6 Full enhancement of tumor after injecting the
contrast 19
3.7 Partial enhancement of tumor after injecting the
contrast 19
3.8 Astrocytmas Low Grade, left T1 and right T2 20
3.9 Glioblastoma (left is T2 image, middle is T1
image and right is T1 with gadolinium) 20
3.10 Classification of segmentation methods 25
4.1 Linear hyper-plane 29
4.2 A non-linear separable case 29
4.3 Linearly separating hyper-planes 30
4.4 (a) hyper-plane with small margin 30
(b) Hyper-plane with large margin 30
4.5 Representation of optimal hyper-plane 31
4.6 Data points and optimal hyper-plane
representation 34
4.7 Representation of slack variables 35

x
4.8 Representation of high dimensional feature space 37
4.9 Transformation of non-linear separator into linear 37
4.10 (a) polynomial classifier (b) RBF 39
5.1 Proposed algorithm 44
5.2 Wavelet Decomposition 49
5.3 2D DWT schematics 49
5.4 Relationship between pair of pixels 52
5.5 5-fold cross validation 54
5.6 Normal brain MR scans 55
5.7 Representation of level 3 wavelet decomposition 56
5.8 Variance curve 57
5.9-5.18 Segmentation results for benign tumor 57-62
5.19-5.28 Segmentation results for malignant tumor 63-68
5.29 Time evaluation at different stages 70

xi
List of Abbreviations

ASM Angular Second Momentum


ADC Apparent Diffusion Coefficient
ANN Artificial Neural Network
CAD Computer Aided Design
CON Contrast
CNN Convolutional Neural Network
CP Cluster Prominence
CS Cluster Shade
CT Computed Tomography
DCT Discrete Cosine Transform
DWT Discrete Wavelet Transform
2D 2-Dimensional
ENT Entropy
EISPACK Eigen System Package
FBB Fast Bounding Box
FDA Food and Drug Administration
GB Giga Byte
GHz Giga Hertz
GLCM Grey Level Co-occurrence Matrix
GRB Gaussian Radial Basis Function
GUI Graphical User Interface
HH, HL High-High, High-Low
IBM International Business Machines
ICA Independent Component Analysis
IDM Inverse Difference Moment

xii
IN Inertia
KNN K-Nearest Neighbours
KSVM Kernel Support Vector Machine
LINPACK Linear System Package
LH Local Homogeneity
LH, LL Low-High, Low-Low
MATLAB MATrix LABoratory
MRI Magnetic Resonance Imaging
MR Magnetic Resonance
NBTF National Brain Tumor Foundation
NMR Nuclear Magnetic Resonance
NSCT Non Sub sampled Transform
PCA Principle Component Analysis
PET Positron Emission Tomography
PNN Probabilistic Neural Network
RAM Random Access Memory
RBF Radial Basis function
SA Simulated Annealing
SE Structuring Element
SGLDM Spatial Gray Level Dependence Matrix
SVM Support Vector Machine
TE, TR Echo time, Repetition Time
TCIA The Cancer Imaging Archive
WHO World Health Organisation
WT Wavelet Transform

xiii

You might also like