Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

INSTITUTE OF STATISTICAL STUDIES AND RESEARCH

CAIRO UNIVERSITY - EGYPT

A Computer-Aided Diagnosis System


in Medical Imaging

by

c Abdalla Mostafa Abdalla

A thesis submitted to the


Institute of Statistical Studies and Research
in partial fulfilment of the requirements for the degree of
Master of Science

in Computer Science

Under the supervision of

Prof. Aboul Elaa Hassanien Asso. Pro. Hesham Ahmed Hefny


Cairo University Cairo University
Faculty of Computers and Information Institute of Statistical Studies and Researches
Department of Information Technology Department of Computer Sciences

2012
I certify that this work has not been accepted in substance for any academic degree
and is not being concurrently submitted in candidate for any other degree.
Any portions of this thesis for which I am indebted to other sources are mentioned
and explicit references are given.

Student:

Abdalla Mostafa Abdalla Zidan

ii
List of Publicatiosns

1) Abdalla Zidan, Neveen I. Ghali, Aboul Ella Hassanien, Hesham Hefny, Jude Hemanth,
Level Set-based CT Liver Computer Aided Diagnosis System, International Journal of
Imaging and Robotics, Volume 9, Issue Number 1, 2013, pp. 26-36.

2) Abdalla Mostafa, Hesham Hefny, Neveen I. Ghali, Aboul Ella Hassanien, Gerald
Schaefer, Evaluating the effects of Image Filters in CT Liver CAD System. IEEE-
EMBS International Conference on Biomedical and Health Informatics (BHI2012). The
Chinese University of Hong Kong, Hong Kong SAR, January 2012, pp 448-451 .

3) Abdalla Mostafa, Aboul Ella Hassanien, Neveen I. Ghali, Hesham Hefny, Level Set-
based Liver Image Segmentation with Watershed and ANN Classifier, 35th International
Conference on Telecommunications and Signal Processing (TSP 2012), Czech Republic,
Prague, 2012, pp. 539-543.

4) Abdalla Mostafa, Neveen I. Ghali, Aboul Ella Hassanien, Hesham Hefny, Level Set-
based CT Liver Image Segmentation with Watershed and Artificial Neural Networks,
The IEEE International Conference on Hybrid Intelligent Systems (HIS2012), Pune, In-
dia, 4-7 Dec. 2012, (accepted).

iii
Acknowledgements

This research would not have been possible without the support of many people. The
author wishes to express his gratitude to his supervisors, Prof. Aboul ella Hassenein
and Assoc. Prof Hesham Hefny who were abundantly helpful and offered invaluable as-
sistance, support and guidance. Deepest gratitude are also due to the members of the
supervisory committee, Prof. Samy Ghonaimy and Prof. Khaled Mostafa Elsayed, with-
out whose knowledge and assistance this study would not have been successful. Special
thanks also to Assoc. Prof. Neveen Ghali who has been always there for support, dis-
cussion, and guidance. And a gratitude for Assoc. Prof. Ashraf Enayet for his medical
support.

iv
Abstract

This thesis aims to design a CAD system for liver diagnosis concentrating on cyst
disease. This is done in three phases, starting with an investigation of filtering tech-
niques to determine the best for image segmentation. Different filtering techniques were
evaluated for this issue, including contrast stretching, sharpening and smoothing tech-
niques. Filtering is used as a preprocessing phase before the segmentation approach to
enhance contrast, remove noise and emphasize certain features as well as connecting ribs
around the liver. The second phase is the image segmentation where liver is separated
from other organs and segmented into a set of regions of interest (ROIs). An algorithm
to connect the ribs is proposed before starting the segmentation. The segmentation ap-
proach combines the level set with watershed approaches used as post segmentation step
to produce a reliable segmentation result. Level set which is a segmentation approach,
is used to separate the whole liver from other organs as stomach, spleen, kidney... etc.
then the separated liver is passed to watershed to extract the ROIs. In the third phase,
features of first order statistics (FOS) and grey-level co-occurrence matrix (GLCM), are
calculated and passed to an artificial neural network (ANN) to be trained and to classify
lesion in ROIs. To evaluate the performance of presented approach, we present tests on
different (Computed Tomography) CT liver images. The experimental results obtained,
show that the overall accuracy offered by the proposed approach for segmenting the whole
liver in CT images is 92.1%, and 88.9% accuracy for neural network classification.

v
Contents

Publications iii

Acknowledgements iv

Abstract v

List of Figures xi

List of Tables xiv

List of Algorithms xv

List of Abbreviations xvi

1 Introduction 1
1.1 Motivations and objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Problem definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Previous work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Thesis organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Main contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Background 8
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Medical background of liver . . . . . . . . . . . . . . . . . . . . . . . . . 9

vi
2.2.1 Liver anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.2 Gall bladder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.3 Blood supply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.4 Liver functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.5 Needle biopsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.6 Liver diseases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.6.1 Hepato-cellular failure . . . . . . . . . . . . . . . . . . . 13
2.2.6.2 Jaundice . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.6.3 Encephalopathy . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.6.4 Hepatic pre-comma and comma . . . . . . . . . . . . . . 15
2.2.6.5 Ascites . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.6.6 Virus hepatitis . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.6.7 Chronic hepatitis . . . . . . . . . . . . . . . . . . . . . . 17
2.2.6.8 Hepatic cirrhosis . . . . . . . . . . . . . . . . . . . . . . 18
2.2.6.9 Hepatocellular carcinoma (HCC) . . . . . . . . . . . . . 18
2.2.6.10 Cyst . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Medical imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Human vision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.2 Imaging using EM spectrum . . . . . . . . . . . . . . . . . . . . . 20
2.3.2.1 Gamma-ray imaging . . . . . . . . . . . . . . . . . . . . 20
2.3.2.2 X-ray imaging . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2.3 MRI imaging . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3.2.4 CT scan imaging . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Mathematical background . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.1 Vector (spatial) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.2 Partial derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4.3 Gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4.4 Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

vii
2.4.5 Curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5 Digital image processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5.1 Image representation . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5.2 Digital imaging concepts . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.3 Relationships between pixels . . . . . . . . . . . . . . . . . . . . . 34
2.6 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.6.1 Region-based techniques . . . . . . . . . . . . . . . . . . . . . . . 36
2.6.1.1 Clustering techniques . . . . . . . . . . . . . . . . . . . 38
2.6.1.2 Split and merge technique . . . . . . . . . . . . . . . . . 43
2.6.1.3 Normalized cuts . . . . . . . . . . . . . . . . . . . . . . 44
2.6.1.4 Thresholding . . . . . . . . . . . . . . . . . . . . . . . . 45
2.6.2 Edge-based techniques . . . . . . . . . . . . . . . . . . . . . . . . 46
2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3 Image preprocessing 49
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2 Filtering concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3 Filtering techniques demonstration . . . . . . . . . . . . . . . . . . . . . 54
3.3.1 Contrast stretching . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.2 Correlation transformation . . . . . . . . . . . . . . . . . . . . . . 54
3.3.3 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3.4 Logarithm transformation . . . . . . . . . . . . . . . . . . . . . . 56
3.3.5 Histogram equalization . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.6 Average (mean) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.7 Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3.8 Gaussian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3.9 Laplacian of gaussian (LoG) . . . . . . . . . . . . . . . . . . . . . 60
3.3.10 Prewitt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.3.11 Sobel filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

viii
3.3.12 Median filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4 Level set and watershed segmentation 65


4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Level set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2.1 Level set overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2.2 Level set implicit functions . . . . . . . . . . . . . . . . . . . . . . 68
4.2.3 Distance function . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2.4 Level Set Methods . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2.4.1 Motion in an externally generated velocity field . . . . . 73
4.2.4.2 Motion involving mean curvature . . . . . . . . . . . . . 78
4.2.4.3 Energy function . . . . . . . . . . . . . . . . . . . . . . . 79
4.3 Watershed segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.3.1 Watershed overview . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.3.2 Watershed techniques . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.3.2.1 Watershed based on flooding . . . . . . . . . . . . . . . 82
4.3.2.2 Watershed based on path-cost minimization . . . . . . . 83
4.3.2.3 Topological watershed . . . . . . . . . . . . . . . . . . . 83
4.3.2.4 Watershed transform based on local condition . . . . . . 84
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

5 Feature extraction and classifications 85


5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2 Feature extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3 Classifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.4 Classification using ANN . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

ix
6 Proposed model 90
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.2 Preprocessing phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.2.1 Connecting ribs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.2.2 Evaluating the proposed filters . . . . . . . . . . . . . . . . . . . . 93
6.3 Level set phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.4 Watershed phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.5 Feature Extraction and classification phase . . . . . . . . . . . . . . . . . 99
6.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

7 Experiments and results 103


7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.2 Image preprocessing phase . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.3 level set and watershed phase . . . . . . . . . . . . . . . . . . . . . . . . 112
7.3.1 Level set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.3.2 Watershed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.4 Feature extraction and classification . . . . . . . . . . . . . . . . . . . . . 125
7.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

8 Conclusion and future work 128


8.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.1.1 Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.1.2 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.1.3 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
8.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

Glossary 131

References 137

x
List of Figures

2.1 Liver Anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10


2.2 The main three slicing movement in liver imaging . . . . . . . . . . . . . 10
2.3 Needle biopsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 Encephalopathy test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Liver cirrhosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Hepatocelular carcinoma in the arterial phase . . . . . . . . . . . . . . . 19
2.7 Different slices of liver and surrounding organs . . . . . . . . . . . . . . . 24
2.8 Different slices of liver and surrounding organs, cont. . . . . . . . . . . . 25
2.9 Vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.10 Image representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.11 Image representation using pixel info tool. . . . . . . . . . . . . . . . . . 31
2.12 Mask values in convolution and correlation . . . . . . . . . . . . . . . . . 32
2.13 Mask sliding in correlation and convolution . . . . . . . . . . . . . . . . . 33
2.14 Adjacency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.15 Different segmentation methods . . . . . . . . . . . . . . . . . . . . . . . 37
2.16 Region Growing starting with seed point . . . . . . . . . . . . . . . . . . 38
2.17 Clustering classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.18 Liver segmentation using different color cluster . . . . . . . . . . . . . . . 40
2.19 Types of Edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.20 Edge-based segmentation of the abdomen . . . . . . . . . . . . . . . . . . 48

3.1 Contrast stretching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

xi
3.2 Correlation filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3 Convolution filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4 Logarithm filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.5 Histogram equalization filter . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.6 Effect of average filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.7 Effect of laplacian filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.8 The effect of gaussian filter . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.9 Effect of laplacian of gaussian . . . . . . . . . . . . . . . . . . . . . . . . 61
3.10 Effect of prewitt filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.11 Effect of sobel filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.12 Effect of Median filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4.1 Zero isocontour of φ(x) = x2 - 1 . . . . . . . . . . . . . . . . . . . . . . . 67


4.2 Zero isocontour of φ(x) = x2 + y 2 = 1 . . . . . . . . . . . . . . . . . . . 67
4.3 Level set contour [Tsai R. et al., 2003] . . . . . . . . . . . . . . . . . . . 68
4.4 Level set narrow band. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.5 Watershed basins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

6.1 The proposed liver CAD system procedure diagram . . . . . . . . . . . . 91


6.2 Contrast stretching to emphasize the ribs feature using the proposed al-
gorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.3 Creaseg application for evaluating different level set methods . . . . . . . 97
6.4 Watershed Form. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.5 Neural Network plotted performance . . . . . . . . . . . . . . . . . . . . 100
6.6 Neural Network application form . . . . . . . . . . . . . . . . . . . . . . 101

7.1 Connecting Ribs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105


7.2 Sample results of some filter techniques . . . . . . . . . . . . . . . . . . . 106
7.3 Comparison of filtering method using SI . . . . . . . . . . . . . . . . . . 111
7.4 Comparison of filtering method using MSE . . . . . . . . . . . . . . . . . 111

xii
7.5 Implementation of Different segmentation methods screen . . . . . . . . . 114
7.6 Different segmentation methods - cont. . . . . . . . . . . . . . . . . . . . 115
7.7 Different segmentation methods - cont. . . . . . . . . . . . . . . . . . . . 116
7.8 Different segmentation methods - cont. . . . . . . . . . . . . . . . . . . . 117
7.9 Different segmentation methods - cont. . . . . . . . . . . . . . . . . . . . 118
7.10 Different segmentation methods - cont. . . . . . . . . . . . . . . . . . . . 119
7.11 Similarity index calculated for CT image cases . . . . . . . . . . . . . . 120
7.12 Comparison between Caselles, region growing and proposed approach . . 121
7.13 Comparison between Caselles, region growing and proposed approach . . 121
7.14 Comparison between 5 level set methods and region growing . . . . . . . 122
7.15 Four different cases for segmented image using level set and watershed . . 124
7.16 Image classification using ANN . . . . . . . . . . . . . . . . . . . . . . . 126

xiii
List of Tables

2.1 Different Slices of liver and surrounding organs . . . . . . . . . . . . . . . 23

7.1 Comparison of filters combination between method (1-6) . . . . . . . . . 107


7.2 Comparison of filters combination between methods(1-6) cont. . . . . . 108
7.3 Comparison of filters combination between methods(7-11) . . . . . . . . 109
7.4 Comparison of filters combination between methods(7-11) cont. . . . . . 110
7.5 Different filters evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.6 Different level set methods evaluation . . . . . . . . . . . . . . . . . . . . 122

xiv
List of Algorithms

4.1 Liver segmentation using level set . . . . . . . . . . . . . . . . . . . . . . 79


4.2 Watershed algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.1 Neural network classification . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.1 Ribs boundary algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2 Creaseg implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

xv
List of Abbreviations
φ Region.
AI Artificial ntelience.
ANN Artificial neural network.
CAD Computer aided diagnosis system.
CT Computed tomography.
EHAA Epidemic hepatitis A antigen.
EM Electromagnetic.
FOS First order statistics.
GLCM Grey-level co-occurance matrix.
HCC Hepatocellular carcinoma.
HCV Hepatitis C virus.
HU Hounsfield unit.
ITK Insight toolkit.
JPG Joint Photographic Experts Group.
KNN K-nearest neighbourhood.
LOG Logarithm.
LoG Laplacian of Gaussian.
MRI Magnetic resonance imaging.
MSE Mean square error.
NN Neural network.
PET Positron Emission Tomography.
R Region.
ROI Region of interest.
SI Similarity index.

xvi
Chapter 1

Introduction

Nowadays, many efforts have been done to support radiologists decision mak-
ing by an efficient computer aided diagnosis (CAD) system. CAD systems may go
around blood tests, fibroscan [Fusiwara S et al., 2007], imaging using Ultrasound, CT
and MRI [Michael Y. et al., 2011]. Every CAD system should consist of some main
phases [Gonzales R. et al., 2004]. Starting with image acquisition, a preprocessing
phase should rise, it may include image registration, rotation, resizing and filtering.
Filtering is an essential step to emphasize features and remove the effect of noise and
distortion. Then, the image is segmented in two different steps, separating liver from
other organs using a level set method, and dividing the extracted liver into regions of
interest using watershed segmentation method. While separating liver requires a smooth-
ing filter that hides the unnecessary details such as arteries, veins and lesions. Extracting
regions of interest (ROIs) requires a sharpening filter to stress the regions edges. Liver
segmentation is the process of separating the anatomical structure of liver from the ab-
domen and extracting ROIs. Intensity is the main key for segmentation, which is used
for determining boundaries of regions and extracting features. Many scales are used
for measuring intensity as HU and JPG. Hounsfield unit (HU)is a quantitative scale for
describing radio-density, that measures the intensity of CT images in the range between
-1500 and 1500 [Michael Y. et al., 2011]. JPG (Compressed image format standardized

1
by the Joint Photographic Experts Group) has the range between 0 and 512, and both
(HU and JPG)could be implemented here [Gonzales R. et al., 2008]. Since liver inten-
sities in CT image is similar to other organs intensities like spleen and muscles, liver
area extraction is convenient for next steps which include segmentation of ROIs and
extracting features. The proposed approach in this paper starts with a smoothed image
to segment the whole liver using level set method, then, pass the segmented liver to wa-
tershed method that segments it into connected ROIs. After that, six types of features
are calculated and trained using an artificial neural network to classify the lesions.

1.1 Motivations and objectives


Liver is an important organ that acts as a carbohydrates and fats store, a controller
of proteins production, a producer of blood clotting factors and a removal of toxins.
So, attention should be paid to liver diseases because of its important effect on the
human’ s health and life. The damage to liver cells causes necrosis, which is the death
of the cells. This is the first step in liver fibrosis, which consists of spots of these
cells. Cirrhosis is the spread of fibrosis in the liver, that moves to a stage where liver
functions are not performed. Cirrhosis is not reversible, it may lead to liver failure or
carcinoma [Feig W. et al, 2004], which means definite death. The early discovery of liver
diseases like cyst, fatty liver lies on the line of recovery stage or at least stopping the
case from getting worse.

This thesis aims to improve the adoption of CAD system in a semi-automatic liver
diagnosis based on CT images.

1.2 Problem definition


Liver is an important organ, performing many functionalities that affect man’s life.
The statistics of liver diseases show that the ratio of virus C infection is 12.8% in Egypt,

2
and the ratio of virus C infection is almost 1.2% in Europe. Also, 130 thousands people
need liver transplantation In Egypt [www.aawsat, 2012].

The early discovery of liver diseases leads to a better treatment and a great chance
for recovery. Late discovery leads to a situation which is not reversible. Needle biopsy
is the golden measure for liver diseases, but it has its limitations and risks. A computer
aided diagnosis (CAD) system is needed to help the specialist to replace the biopsy
and discover the lesions early. Many challenges face CAD systems, it rises from the
complexity of the three phases of image manipulation for diagnosis and getting high
accuracy rates of determining the lesion. Aiming for accuracy and low cost for imaging,
the research is concerned in making the automatic diagnosis available in low cost rates.

1.3 Previous work


This thesis is not intending to reinvent the wheel. It is a trial to feel the environ-
ment, and explore the field of image processing, to be a base for launching to a future
little contribution to help the community. Research in medical imaging concentrates
primarily on utilizing computer facilities into the real world of medicine to improve all
areas of the field of diagnosis and treatment. Mathworks has developed Matlab as an
impressive tool which is very helpful for mathematical implementation, image segmenta-
tion and analysis. Also, the National Library of Medicine (NLM) develop and support a
tool for segmentation and registration named insight toolkit (ITK) [www.itk.org, 2012].
ITK is an open-source, cross-platform system that provides developers with an extensive
suite of software tools for image analysis.

Also, there are lots of previous efforts of many researches, working in image pro-
cessing and medical diagnosis. Segmentation using level set methods was first raised by
Stanley Osher in 1987 [Osher S. et al., 2003, Fedkiw R. et al., 2003]. Later on, numer-
ous approaches have been proposed to improve the accuracy of segmentation, as upwind

3
differencing, hamilton-jacobi ENO [Harten A. et al., 1987, Osher S., 1988], weighted
hamilton-jacobi (WENO)
[Liu X. et al., 1996], TVD Runge-Kutta [Osher S., 1988]. Chan and Vese proposed a
novel model that combines the Mumford-Shahs functional and level set methods
[Osher S. et al., 1988], which can handle curves, surfaces with topological changes easily.
Beside these models, [Kass M. et al., 1987] proposed the classic snake model for segmen-
tation, [Malladi R., 1995] developed an active contour model, and [Caselles V. et al., 1997]
developed their geodesic active contour models. Tony Chan implemented the seg-
mentation using a shape prior method [Chan T. et al., 2005]. Cao et al. developed
a method for extracting liver fibrosis in ultrasound images [Gui-taao C. et al., 2005].
Eva Rikxoort used a k-NN voxel classifier and atlas matching for liver segmentation
[Rikxoort E. et al., 2007]. E. Hodneland combined the usage of the four colour the-
orem, level set and watershed methods for segmentation [Hodneland E. et al., 2009].
Jae-Hong proposed a method that extracts the lesion by means of a Fuzzy c-Means
clustering technique, which can determine the threshold regardless of a changing in-
tensity [Hong J. et al., 2001]. E-Liang Chen et al. offered a CT liver image diagnostic
classification system which automatically find the liver boundary, and a neural network
liver classifier uses specially designed feature descriptors to distinguish normal liver, two
types of liver tumors, hepatoma and hemangioma [Chen E. et al 1998]. Stavroula et al.
extracted five distinct sets of texture features. Two different classifiers were used. The
first one consists of five Neural Networks (NNs), each using as input, either one of the
computed texture feature sets or its reduced version after feature selection. The second
ensemble of classifiers was generated by combining five different types of primary classi-
fiers, two NNs, and three k-nearest neighbour classifiers. The decision of each ensemble
of classifiers was extracted by applying voting schemes[Mougiakakou S. et al., 2006].

You might also like