ACKNOWLEDGEMENT

i
ABSTRACT
In medical field, automatic brain diagnosis is very crucial task. It leads to automatic detection of

abnormalities in human brain. The word abnormality means an existence of unwanted tissue in

body that is not vanished with time. MRI imaging technique is most popular as they provide

complete detail of internal structure of body which cannot be seen in other medical imaging

techniques. It also provides good contrast between soft tissues of the body without making use of

any ionizing radiation.

In this work, MRI images of brain have been taken for research. The aim of this work is

automatic detection and classification of human brain abnormalities. Brain diagnosis is done by

extracting some relevant information from images known as features. There are different types of

features in images which can be extracted by using particular extraction techniques such as PCA

and GLCM. This extracted information is sufficient to represent the whole image. This process is

called as feature extraction. This extracted information (features) is further used to classify

between normal and abnormal images depending upon the range of features extracted along with

the type of existing abnormality by using particular classification technique SVM; which is

based upon machine learning. This process is called as feature classification.

This work is done for automatic detection and classification of images to reduce the human

intervention by classifying between normal and abnormal images. Sometimes, it happens that

operator assisted classification leads to cause misclassification because of large amount of data is

available. With the help of particular classification; data can be accurately classified. Abnormal

images are further categorized into the type of existing abnormalities such as tumor, bleed,

trauma, clot, acute-infarct depending upon the values of features extracted. But the final decision

is made after the consultation of doctor.

ii
Along with medical images this procedure is repeated for natural images also; to check the

performance of above mentioned feature extraction and classification techniques. According to

experimental results, GLCM gives high accuracy for both natural as well as medical images.

iii
ORGANIZATION OF THESIS
The first chapter introduces the computer vision system in image processing with image

acquisition, processing and feature extraction. Along with it also includes the basic concepts of

feature extraction and classification with their techniques in brief. The second chapter includes

the literature survey which depicts the work done in this area. The third chapter is based on

biomedical imaging technique, Magnetic resonance Imaging along with its working principle and

scanning system with its applications, advantages and disadvantages. The fourth chapter

discusses about the mathematical background of PCA and SVM with basic concepts. The fifth

chapter discusses the algorithms used in this thesis in detail. The sixth chapter describes learning

methods, its types along with classification and clustering concept with SVM. The seventh

chapter shows the methodology adopted in this research for feature extraction and classification

along with flow chart. The eighth chapter discusses about results obtained from images and lastly

detects whether image is normal or abnormal.

iv
TABLE OF CONTENTS

Contents Page Number

Acknowledgement i

Abstract ii-iii

Organization of Thesis iv

Table of Contents v-ix

List of Abbreviations x

List of Figures xi

List of Tables xiii

CHAPTER-1 INTRODUCTION 1-9

1.1 Computer Vision System 1

1.1.1 Image Acquisition 1

1.1.2 Pre-Processing 2

1.1.3 Feature Extraction 2

1.2 Image Processing 2

1.2.1 Edges 3

1.2.2 Corners 3

v
 1.2.3 Region of Interest 3

1.3 Feature extraction 3

1.3.1 General 4

1.3.2 Feature extraction techniques 4

1.3.2.1 Gray Level Co-occurrence Matrix (GLCM) 5

1.3.2.2 Principal Component Analysis (PCA) 5

1.3.2.3 Independent Component Analysis (ICA) 5

1.3.2.4 Wavelet Transform 6

1.3.2.5 Linear Dependent Analysis (LDA) 6

1.4 Feature Classification 7

1.4.1 Feature Classification Techniques 7

1.4.1.1 Support Vector Machine (SVM) 7

1.4.1.2 K-Nearest Neighbor (K-NN) 8

1.4.1.3 Perceptron Model 8

1.4.1.4 Decision Tree 9

1.4.1.5 Naïve Bayes 9

CHAPTER-2: LITERATURE REVIEW 10-15

CHAPTER-3: MAGNETIC RESONANCE IMAGING (MRI) 16-23

3.1 Introduction 16

3.2 Magnetic Resonance Imaging (MRI) 16

3.3 Working of MRI 17

3.4 Image Contrast and Contrast Enhancement 19

3.5 Basic MRI Scans 19

3.5.1 T1-Weighted MRI 20

3.5.2 T2- Weighted MRI 20

vi
 3.5.3 T2*- Weighted MRI 20

3.5.4 Fluid Attenuated Inversion Recovery (FLAIR) 21

3.5.5 Spin Density Weighted MRI 21

3.6 Specialized MRI Scans 21

3.6.1Diffusion MRI 21

3.7 Applications of MRI 21

3.8 Advantages of MRI 22

3.9 Disadvantages of MRI 22

CHAPTER 4 BACKGROUND AND RELATED WORK 24-38

4.1 Introduction 24

4.2 Statistics 24

4.3 Matrix Algebra 28

4.3.1 Eigenvector and Eigenvalues 29

4.4 Geometry 31

4.4.1 Hyperplane 31

4.4.2 Distance from Point to Line 33

CHAPTER 4 FEATURE EXTRACTION TECHNIQUES 39-51

5.1 Image Texture 39

5.1.1 Structured approach 39

5.1.2 Statistical approach 39

5.2 Gray Level Co-occurrence Matrix (GLCM) 39

5.2.1 Creating Gray Level Co-occurrence Matrix 40

5.3 Principal Component Analysis (PCA) 42

5.3.1 Characteristics of Principal Components 44

5.3.2 Methodology 45

vii
 5.3.3 Mathematical example of PCA 46

5.3.4 Graphical representation of PCA 48

5.4 Applications of PCA 49

5.5 Advantages of PCA 49

5.6 Disadvantages of PCA 49

5.7 Features extracted from images 50

CHAPTER 6 MACHINE LEARNING 52-84

6.1 Introduction 52

6.1.1 Machine Learning Algorithms 52

6.2 Classifier (An-Introduction) 54

6.3 Classification/Clustering 54

6.3.1 Clustering 55

6.3.2 K-Means Clustering 55

6.3.3 Classification 55

6.4 Hyperplane with classifier 57

6.5 Support Vector Machine (SVM) 58

6.5.1 Linearly Separable Binary Classification 60

6.5.2 Linear Non-Separable Classification 66

6.5.3 Non- Linear SVM 72

6.5.4 Kernel functions 74

6.5.5 Kernel methods 76

6.5.6 Kernel Types 83

CHAPTER 7 RESEARCH METHODOLOGY 85-90

7.1 Introduction 85

7.2 MATLAB Software 85

viii
7.3 Flow chart 87

7.4 Algorithm for extraction and classification of features from images using GLCM, PCA
and SVM 88

CHAPTER 8 RESULTS & DISCUSSIONS 91-134

8.1 Results for natural images using PCA and GLCM for training phase 92

8.1.1 Features extracted from natural images using PCA and GLCM for training phase 95

8.1.2 Classification results of natural images using PCA and SVM for test phase 97

8.1.3 Classification results of natural images using GLCM and SVM for test phase 98

8.2 Graphical Performance of PCA and GLCM on natural images using SVM 99

8.3 Results for Brain MRI images using PCA and GLCM for training phase 105

8.4 Results for Brain MRI images using PCA and GLCM with SVM for test phase 118

8.4.1 Brain MRI images for test phase

8.4.2 Classification results of Brain MRI images using PCA and SVM for test phase 123

8.4.3 Classification results of Brain MRI images using GLCM and SVM for test phase 125

8.5 Graphical Performance of PCA and GLCM for Brain MRI images using SVM 127

CONCLUSION 135

FUTURE WORK 137

REFERENCES 138

ix
LIST OF ABBREVIATIONS
AI Artificial Intelligence

CSF Cerebrospinal Fluid

CT Computed Tomography

EEG Electroencephalograph

FLAIR Fluid Attenuated Inversion Recovery

fMRI Functional Magnetic Resonance Imaging

GLCM Gray-Level Co-occurrence Matrix

ICA Independent Component Analysis

KNN K-Nearest Neighbor

LDA Linear Dependent Analysis

MATLAB Matrix Laboratory

MEG Magneto Encephalography

MLP Multi Layer Perceptron

MRI Magnetic Resonance Imaging

MRT Magnetic Resonance Tomography

NMRI Nuclear Magnetic Resonance Imaging

PCs Principal Components

PCA Principal Component Analysis

PET Positron Emission Tomograph

RBF Radial Basis Function

RF Radio Frequency

RGB Red Green Blue

ROI Region of interest

SVM Support Vector Machine

x
LIST OF FIGURES
Figure Page Number
Figure 3.1: MRI image of Brain 17

Figure 3.2: MRI Scanner 18

Figure 3.3: Different types of MRI Images 19-20

Figure 4.1: 1D Hyperplane 32

Figure 4.2: 2D Plane with Straight Line 34

Figure 4.3: P Point at a distance PQ from line DE 34

Figure 4.4: PQ Distance from Line 35

Figure 4.5: RS new line with same distance parallel to PQ 36

Figure 5.1: Test Image 41

Figure 5.2: Pixel values corresponding to Test Image 41

Figure 5.3: Gray Level Concurrence Matrix 41

Figure 5.5: Variance of data points 43

Figure 5.6: Plot of Principal Components 49

Figure 6.1: Block diagram of classifier 54

Figure 6.2: Clustering and Classification Process 57

Figure 6.3: General classifier 58

Figure 6.4: Possible hyperplanes 59

Figure 6.5: Optimal hyperplane with margin 60

Figure 6.6: SVM with linearly separable data points 61

Figure 6.7a: Margin with narrow width 66

Figure 6.7b: Margin with widest width 67

Figure 6.8: SVM with not fully linearly separable data with 1 misclassified data point 67

Figure 6.9a: SVM with not fully linearly separable data points 68

xi
Figure 6.9b: SVM with not fully linearly separable data points 69

Figure 6.10: Hard Margin 71

Figure 6.11: Soft Margin 71

Figure 6.12: Data points non-linear separable 73

Figure 6.13: Kernels mapping idea 73

Figure 6.14: XOR gate data points in 2D 74

Figure 6.15: XOR data points in 3D 75

Figure 6.16: XOR data points classified without Kernel Function 76

Figure 6.17: XOR data points classified with kernel function 76

Figure 6.18: Data points on 1D space 77

Figure 6.19: Data Points on 2D Spaces 78

Figure 6.20: Hyperplane with maximum margin in feature space 82

Figure 7.1: Flow chart for the determination of type of abnormality in image 87

Figure 8.1: PCA with Linear Kernel Function for Natural Images 100

Figure 8.2: GLCM with Linear Kernel Function for Natural Images 101

Figure 8.3: PCA with Quadratic Kernel Function for Natural Images 102

Figure 8.4: GLCM with Quadratic Kernel Function for Natural Images 103

Figure 8.5: PCA with RBF Kernel Function for Natural Images 104

Figure 8.6: GLCM with RBF Kernel Function for Natural Images 105

Figure 8.7: PCA with Linear Kernel Function for MRI Images 128

Figure 8.8: GLCM with Linear Kernel Function for MRI Images 129

Figure 8.9: PCA with Quadratic Kernel Function for MRI Images 130

Figure 8.10: GLCM with Quadratic Kernel Function for MRI Images 131

Figure 8.11: PCA with RBF Kernel Function for MRI Images 132

Figure 8.12: GLCM with RBF Kernel Function for MRI Images 133

xii
LIST OF TABLES
Tables Page Number
Table 6.1 Regularization Parameter 70

Table 8.1: Names assigned to natural images 92

Table 8.2: Feature Extracted from Natural Images using PCA and GLCM for Training Phase 95

Table 8.3: Results of Natural Images using PCA and SVM for Test Phase 97

Table 8.4: Results of Natural Images using GLCM and SVM for Test Phase 98

Table 8.5: Names assigned to Brain MRI Images for training phase 106

Table 8.6: Features Extracted from Brain MRI Images using PCA and GLCM for Training
Phase 111

Table 8.7: Names assigned to Brain MRI images for test phase 118

Table 8.8: Results of Brain MRI Images using PCA and SVM for Test Phase 123

Table 8.9: Results of Brain MRI Images using PCA and SVM for Test Phase 125

Table 8.10: Combined results for Natural Images using PCA and GLCM 134

Table 8.11: Combined results for Brain MRI Images using PCA and GLCM 134

xiii