automatic detection of diabetic retinopathy … detection of diabetic... · automatic detection of...
TRANSCRIPT
AUTOMATIC DETECTION OF DIABETIC RETINOPATHY INCLUDING NEOVASCULARIZATION BASED ON
MORPHOLOGICAL OPERATIONS
~
Siti Syafinah Binti Ahmad Hassan
Master of Engineering 2012
'usa. KhiOmal Ma mal A a.emit-. UNIVERSm MALAVSIA SARAWAK
(
AUTOMATIC DETECTION OF DIABETIC RETINOPATHY INCLUDING NEOV ASCULARIZATION BASED ON
MORPHOLOGICAL OPERATIONS
SITI SYAFINAH BINTI AHMAD HASSAN
A thesis submitted in fulfillment of requirements for the degree of Master of Engineering
"1.
F acuity of Engineering UNIVERSITI MALAYSIA SARAW AK
2011
,......... I
DECLARATION
I would like to declare that this dissertation is my original writing, except the data, the notes
and fact that already stated with its sources and origins.
~-
SIT! SY AFINAH BINT! AHMAD HASSAN 850909-13-5830
:For he{ovedMOM and1J.Jt1J
11
I
ACKNOWLEDGEMENTS
I am heartily thankful to my supervisor, Ir. David Bong Boon Liang for his encouragement,
guidance and support from the initial. to the final level that enabled me to develop the project.
Special thanks are given to my co-supervisor which also acts as consultant ophthalmologist
Dr. Maflika Premsenthil for her kind supervision and provision of medical expertise. I am
also thankful to Local Hospital and all the publically available databases (DRIVE, STARE,
" DiaretdbO arrd Messidor program) for providing the diabetic retinal images and clinical
infonnation. It is a pleasure to thank Ministry of Science, Technology and Innovation,
Malaysia for the financial support on this project.
Lastly, I offer my regards and blessings to all of those who supported me in any respect
during the completion of the project especially my parents for their loving support and
encouragement in so many ways. I am sure the success of this project would make them
proud. I cannot thank enough my special friend for his love and support, without which I
would never have completed this work.
11\
I
ABSTRACT
("Diabetic retinopathy is a widely spread eye disease caused by diabetes complication.
Screening to detect retinopathy disease can lead to successful treatments in preventing
blindness especially at early stages. An automated decision support system for the purpose of
detecting and classifying retinal abnormalities is carried out mainly using an image processing
methods as presented in this thesis. The retinal images are automatically analyzed in term of
pixel-based and region-based diagnostics accuracies after compared with ophthalmologist's
hand-trut~ An adjusted morphology-based, thresholding and mathematical-based pixel
segmentation methods are developed to segment the bright lesions from background and to
distinguish from other retinal feature especially optic disc. Gradient classifier has been used to
distinguish hard exudates and cotton wool spot from bright lesions segmentation result. The
preliminary pixel-based hard exudates and pixel-base cotton wool spot analysis are used to
support the fact that development of a reliable retinal abnormalities identification system is
feasible. There are small dark lesions and large dark lesions detection methods presented in
this thesis. Image enhancement, image restoration, morphology operator, thresholding and
compactness properties techniques have been used in development of automatic dark lesions
detection system. Lastly, neovascularization lesion identification is still a new study in
automatic detection of diabetic retinopathy. Detection of neovascularization is important since
it signifies the disease has reaches a vision-threatening phase. Therefore, image normalization,
morphology-based operator, Gaussian filtering and thresholding techniques are used in
developing of neovascularization detection. Moreover, a function matrix box-based on
predefined criteria of neovascularization has been used in order to classify the
neovascularization from natural blood vessel. The developed method is tested on a set of 303
IV
l
images from different database sources to classify the images as abnormal or normal images.
The proposed method managed to achieved 90.29% score for abnormal images and 100%
score for normal images. Result after testing shows 85.39% sensitivity and 94.59% specificity
for bright lesions, 70.68% sensitivity and 99.22% specificity for dark lesions and 63.9%
sensitivity and 89.4% specificity for neovascularization. Further detail of lesions shows
61.24~ sensitivity and 98.43% specificity for hard exudates, 48.62% sensitivity and 98.75%
specificity for cotton wool spots, 27.41% sensitivity and 99.94% specificity for
microaneurysms and 67.93% sensitivity and 99.70% specificity for haemorrhages.
Keywords: Diabetic retinopathy, dark lesions, bright lesions, hard exudates, cotton wool
spots, neovascularization, image processing.
v
II,....... ,
ABSTRAK
Diabetik retinopati adalah penyakit mata yang semakin berleluasa djsebabkan oleh komplikasi
diabetes.; Pengujian untuk mengesan penyakit retinopati membolehkan rawatan awal
dilakukan untuk mengelak kerosakan penglihatan pada peringkat awal. Sistem sokongan
automatik ini bertujuan mengesan dan mengkategorikan abnormal retina menggunakan
kaedah pemprosesan imej seperti diterangkan dalam tesis ini. Imej retina dianalisa secara
automatik dari segi ketepatan diagnosis kawasan dan piksel selepas perbandingan dengan
lakaran oftalmologi. Pengolahan berasaskan morfologi, proses pembahagian dan kaedah
matematik berasaskan segmen piksel telah dibina untuk memisahkan bright lesions daripada
latar belakang dan membezakan dari ciri-ciri retina yang lain terutamanya cakera optik.
Pengelasan berdasarkan kecerunan telah digunakan untuk membezakan hard exudates dan
cotton wool spot daripada keputusan segmentasi bright lesions. Analisa asas berdasarkan
piksel untuk hard exudates dan cotton wool spot dapat digunakan untuk menyokong
kenyataan bahawa pembangunan sistem pengesanan ketidaknormalan retina yang mempunyai
keboleharapan yang tinggi adalah munasabah. Terdapat kaedah pengesanan dark lesions yang
keeil dan besar ditunjukkan dalam tesis ini. Penambah baik imej, pembaharuan imej, operasi
morfologi, proses pembahagian dan teknik ciri-ciri kepadatan telah digunakan dalam
membangunkan sistem pengesanan automatik dark lesions. Akhir sekali, pengesanan simtom
neovascularization merupakan satu kajian yang masih baru dalam pengesanan automatik
diabetik retinopati. Pengesanan neovascularization sangat penting kerana ia menunjukkan
penyakit sudah mencapai fasa rosak penglihatan. Oleh itu, penyeragaman imej operasi asas
morfoiogi, penapisan Gaussian dan proses pembahagian digunakan dalam membangunkan
sistem pengesanan ciri-ciri neovascularization. Selanjutnya, fungsi kotak matrik berasaskan
VI
l
kriteria yang telah ditentukan digunakan untuk pengkelasan neovascularization dari salur
darah yang normal. Kaedah yang dibangunkan telah diuji dengan menggunakan 303 imej dari
pelbagai sumber untuk pengkelasan imej normal dan imej tidak normal. Kaedah ini berjaya
mencapai keputusan 90.29% markah bagi imej tidak normal dan 100% markah bagi imej
normal. Keputusan selepas pengujian kaedah menunjukkan 85.39% sensitif dan 94.59%
spesifik bagi bright lesions, 70.68% sensitif dan 99.22% spesifik bagj dark lesions dan
58.95% sensitif dan 88.03% spesifik bagi neovascularization. Lesions yang lebih terperinci
menunujukkan 61 .24% sensitif dan 98.43% spesifik bagi hard exudates, 48.62% sensitif dan
98.75% spesifik bagi cotton wool spots, 27.41 % sensitif dan 99.94% spesifik bagi
microaneurysms dan 67.93% sensitif dan 99.70 spesifik bagi haemorrhages.
Kata kunci: Diabetik retinopati, dark lesions, bright lesions, hard exudates, cotton wool
spots, neovascularization, pemprosesan imej .
VII
Pusat Kltidmat Maldumat Akademik UNlVERSm MALAYSIA SARAWAJ(
TABLE OF CONTENTS
Page
ACKNOWLEDGEMENTS III
ABSTRACT IV
ABSTRAK VI
TABLE OF CONTENTS viii
LIST OF FIGURES x
LIST OF TABLES Xlll
LIST OF ABBREVIATIONS XIV
CHAPTER 1: INTRODUCTION 1. ] Background 1 1.2 Problem statements 3 1.3 Scope and objectives 4 1.4 Research overview 5 1.5 Thesis structure 6
CHAPTER 2: DETECTION OF RETINAL ABNORMALITIES USING IMAGE PROCESSING 2.1 Introduction 7 2.2 Structure of the eye 7 2.3 Diabetic retinopathy background 9
2.3.1 Diabetic retinopathy complications 10 2.3.2 Diabetic retinopathy risk factor and preventions 15 2.3.3 Screening for diabetic retinopathy 16
2.4 Computer vision fundamental 19 2.4.1 Image acquisition 21 2.4.2 Image pre-processing 21 2.4.3 Image segmentation 24
2.5 Computer vision in related diabetic retinopathy studies 27
CHAPTER 3: METHODOLOGY 3.1 Introduction 34 3.2 Image acquisition 36 3.3 Bright lesions detection 38
3.3.1 Pre-processing for retinal image 40
VIII
l 3.3.2 Bright lesions segmentation 42 3.3.3 Bright lesions classification 58 3.3.4 Bright lesions quantification 64
3.4 Dark lesions detection 65 3.4.1 Dark lesions segmentation 68 3.4.2 Dark lesions classification 77 3.4.3 Dark lesions quantification 83
3.5 Neovascularization detection 84 3.5.1 Neovascularization segmentation 88 3.5.2 Neovascularization classification 88
3.6 Abnormal and normal classification 90
CHAPTER 4: RESULTS AND DISCUSSION 4.1 Introduction 92 4.2 Sensitivity and Specificity analysis 92
4.2.1 Bright lesions 93 4.2.2 Dark lesions 98 4.2.3 Neovascularization lesions 101
4.3 Receiver operating characteristic (ROC) Graph 103 4.3.1 Bright lesions 104 4.3.2 Dark lesions 107 4.3.3 Neovascularizations 110
4.4 Classification of normal and abnormal images 112 4.5 Comparative study 112 4.6 Discussion 118
CHAPTER 5: CONCLUSIONS AND RECOMMENDATIONS 5.1 Introduction 121 5.2 Conclusions 121 5.3 Recommendations 126
REFERENCES
APPENDIX A: Bright lesions detection - SN and SP for 256 test images 135 139 151 155 167 167 172 179 196 204
APPENDIXB: Test result of automatic bright lesion detection APPENDIXC: Dark lesions detection - SN and SP for 256 test images APPENDIXD: Test result of automatic dark lesion detection APPENDIX E: Neo detection - SN and SP for 9 test images APPENDIX F: Test result of automatic neo detection APPENDIXG: Images summary APPENDIXH: Algorithm for the proposed research APPENDIX I: Journal of Digital Imaging Springer (2011) APPENDIXJ: National Conference on Engineering and
APPENDIXK:
127
Technopreneurship (NCET 2010) Proceedings of EN CON 2010 209
IX
-- -
I' ,...
Figure
Figure 1.1: 1
Figure 2.1:
Figure 2.2: Figure 2.3: Figure 2.3: Figure 2.4: Figure 2.5: Figure 2.6: Figure 2.7: Figure 2.8: Figure 2.9:
Figure 2.10: Figure 2.11 : Figure 2.12: Figure 3.1: Figure 3.2: Figure 3.3:
Figure 3.4:
Figure 3.5:
Figure 3.6: Figure 3.7:
I' Figure 3.8:
Figure 3.9:
Figure 3.10: Figure 3.11:
Figure 3.12:
LIST OF FIGURES
Title Page
Outline of the proposed system for automatic detection of retinal 6 abnormalities Structure of eye; (a) cross sectional view of human eye; (b) A 8 digital photograph of the retina or fundus image Microaneurysms 13 (a) Blot and dot haemorrhages; 13 (b) Flame haemorrhages 14 Hard exudates 14 Cotton wool spot 15 Neovascularization 15 A 2x2 contingency table 19 Fundamental steps in digital image processing 20 Example of spatial filtering; (a) Original image; (b) Result of noise 22 reduction with 3x3 mean filter; (c) Result of noise reduction with 3x3 median filter Gray - scale transformations 23 The Dilation operation 26 The Erosion operation 26 The main stage in automatic detection of retinal abnormalities 35 Screening of diabetic retinopathy using fundus camera 38 The outline of the proposed system for automatic identification of 39 retinal bright lesions. The process for bright retinal lesions segmentation of the proposed 43 system (a) Green channel image; (b) Morphology dilation image; (c) 46 Morphology erosion for image (b) Window size 5x5 47 Four orientations in window; (a) 0 degree; (b) 45 degree; (c) 90 48 degree; (d) 135 degree (a) mean difference value = I; (b) mean difference value = 3; (c) 50 mean difference value = 5; (d) mean difference value = 7; (e) mean difference value = 10 (a) Identification of midpoint; (b) After thinning; (c) After flood-fill 51 operation; (d) After morphological opening operation The process for optic disc localization of the proposed system 53 The procedure for bright retinal feature extraction and region 58 classification of the proposed system (a) Reconstructed image of the pre-processed image; (b) Bright 60 lesions region obtained after thresholding the result image of subtraction between (a) and pre-processed image
x
~~------~________________________________________________________________~;i
I r
Figure 3.13:
Figure 3.14: Figure 3.15:
Figure 3.16:
Figure 3.17: Figure 3.'18:
Figure 3.19: Figure 3.20: Figure 3.21: Figure 3.22 Figure 3.23: Figure 3.24:
Figure 3.25:
Figure 3.26:
Figure 3.27: Figure 3.28: Figure 3.29: Figure 3.30:
Figure 3.31: Figure 3.32:
Figure 3.33:
Figure 3.34:
Figure 3.35:
Figure 3.36:
Figure 3.37: Figure 4.1:
Figure 4.2:
Figure 4.3:
Figure 4.4:
The procedure for advance bright lesions classification to 61 distinguish between hard exudates and cotton wool spots of the proposed system (a) Hard exudates; (b) Cotton wool spots 63 The procedure for quantifying number of hard exudates and cotton 64 wool spots that signify occurrences of Diabetic Retinopathy The outline of the proposed system for automatic identification of 67 retinal dark lesions The procedure for dark lesions segmentation of the proposed system 69 Block diagram of dark expose function for dark segmentation Contrast Enhancement; (a) , imadjust'; (b) , adjcontrast' Dilation Erosion 'dark expose' result image Image subtraction
lesions 70
71 72 73 73 74
Block diagram of blood vessel extraction function for dark lesion 76 segmentation (a) Thresholding; (b)Morphology open filter; (c) Compactness 77 classifier Flow chart for dark retinal feature extraction and reglOn 79 classification of the proposed system Morphology fill; (a) Image subtraction; (b) Small dark lesions 80 (a) Early subtraction; (b) After morphology open 81 Fovea 82 The procedure for advance dark lesions classification to distinguish 82 between microaneurysms and haemorrhages of the proposed system Total dark lesion 83 The procedure for quantifying number of microaneurysms and 84 haemorrhages that signify occurrences of Diabetic Retinopathy The outline of the proposed system for automatic identification of 85 retinal neovascularization lesions (a) Image after applying pre - processing operation; (b) Image after 87 applying match - based filter; (b) Contrast stretched image; (b) Image after mUltiple morphology dilation and erosion; (b) Image subtraction between (c) and (d); (f) Morphology opening The procedure for Neovascularization segmentation of the proposed 88 system The procedure for Neovascularization feature extraction and region 89 classification of the proposed system Neovascularization classification result 90 Samples of results comparing the bright lesions reglOn with 97 ophthalmologist's hand-truth Samples of results comparmg the dark lesions region with 99 ophthalmologist's hand-truth Samples of results comparing the neovascularizations region with 102 ophthalmologist's hand-truth Comparing ROC Curves 104
XI
I Figure 4.5: ROC curve for bright lesion detection 107
Figure 4.6: ROC curve for dark lesion detection 110
Figure 4.7: ROC curve for Neovascu1arization detection 111
XII
I ~
LIST OF TABLES
Table Title Page
Table 2.1: I
International clinical Diabetic Retinopathy Disease Severity scale 11 Table 2.2: Screening follow up schedule of Diabetic Retinopathy stage 17 Table 4.1: The result of bright lesion detection in 47 retinal image that used in 95
developing the proposed algorithm Table 4.2: The result of dark lesions detection in 47 retinal image that used in 100
developing the proposed algorithm Table 4.3: The result of neo detection in 3 retinal image that were used in 101
developing the proposed algorithm Table 4.4: Table of TP and FP used to construct the ROC curve of bright 106
lesions detection generated by online software Table 4.5: Table of TP and FP used to construct the ROC curve of dark lesions 109
detection generated by online software Table 4.6: Table of TP and FP used to construct the ROC curve of 110
neovascularization detection generated by online software Table 4.7: Comparison of automated detection of diabetic retinopathy 117 Table 4.8: Sensitivity and specificity result in developing the proposed system 119 Table 4.9: Sensitivity and specificity result in testing the proposed system 119 Table 4.10: Summary result of abnormal and normal image 119
XliI
DR
NPDR
PDR
IRMA
SN
SP
TP
IN
FN
FP
ROC
FOV
NEO
LIST OF ABBREVIATIONS
Diabetic Retinopathy
Non.proliferative Diabetic Retinopathy
Proliferative Diabetic Retinopathy
Intra Retinal Microvascular Abnonnalities
Sensitivity
Specificity
True Positive
True Negative
False Negative
False Positive
Receiver Operating Characteristic
Field of view
Neovascularization
XIV
CHAPTERl
I
INTRODUCTION
1.1 BACKGROUND
This thesis explores the identification of diabetic eye disease in digital images of human
eye. Image processing techniques are introduced to detect and classify picture element of
digital retinal images into diabetic eye lesion classes.
The rising cases of diabetes mellitus give great challenges to public health care. Diabetes
prevalence rate in Malaysia has risen much faster than expected, almost doubling in
magnitude over the last decade (Zanariah, H. et at. 2006). International Diabetes Federation
(lDF) predicted that by the year 2025, the South East Asian Region would have the highest
prevalence of diabetes. World Health Organization (WHO) also has estimated that by the year
2030, Malaysia would have a total of 2.48 million people with diabetes (Zanariah, H. et al.
2006).
The First National Health and Morbidity Survey (NHMS I) conducted in 1986 had
reported a prevalence of diabetes of 6.3% and in the Second National Health and Morbidity
Survey in 1996 (Maimunah, A.H, 1996), this had risen to 8.3%. The Third National Health
and Morbidity Survey (NHMS III) in 2006 shows the prevalence of diabetes is 14.9% among
the population aged 30 years and above. Most of the patients are not aware of the diabetic
progression until it gets worse. Therefore, late treatment of diabetic disease will consequently
leads to various complications especially retinopathy.
Diabetic retinopathy (DR) is the most common cause of blindness in adults of working
age in Malaysia. The diabetic retinopathy stage varies with the duration of disease. At
diagnosis, less than 5% of the patients have retinopathy while after 10 years the prevalence
rate has risen to 40-50%. After 20 years, almost all patients with type I diabetes and more than
60% patients with type II diabetes have some degree of retinopathy (Jaudin, R. 2006).
The National Eye Survey that was carried out in 1996 showed that the prevalence of
diagnosed diabetes in Malaysia aged 50 years and above was estimated to be 10.3% or about
200,000 people and 3.5% of them might have diabetic retinopathy. Thus, about 7,300 people
aged 50 years and above were estimated to have diabetic retinopathy in Malaysia. Diabetic
retinopathy has become an increasingly important cause of blindness. Nevertheless, vision
loss can be prevented from early detection of diabetic retinopathy and monitor the existing
retinopathy with regular examination (Jaudin, R. 2006).
Most of the patients are being treated at the primary care level-general practitioner clinics,
public out-patient clinics and health centers. The huge number of patients at these primary
care clinics is in sharp contrast to the limited number of expert ophthalmologists that need to
sereen and review the retinal images of the patients. A study by Javitt, J.C. (1989) and Jaudin,
It (2006) showed that the high cost of screening and shortage of medical professionals are the
fictors that hamper patients from taking regular screening.
2
With the advancement of digital image processing technologies, the techniques can be
applied to analyze retinal images in the absence of expert ophthalmologists. The automatic
diagnosis and analysis techniques can help ophthalmologists in diabetic screening for
detecting the symptoms faster, more easily and reduce ophthalmologist workload especially in
developing country where the number of experts is not sufficient to deal with the large
number of diabetic retinopathy patients.
The research work in this thesis aims to automatically diagnose abnormalities in retinal
images and further analyze the abnormalities. This automated process would speed up
diagnosis of retinopathy, thus cases which require urgent attention could be referred
immediately to an expert for further evaluations.
1.2 PROBLEM STATEMENTS
A regular eye examination done by an expert ophthalmologist is a manual way in
diagnosing and quantifying retinal abnormalities of each patient. Consequently, this manual
method increases the expert workload and they will be inclined to spend less time to diagnose
one retinal image, thus higher tendency of error in diagnosis. Most of the previous researches;
I
Walter, T. et al. (2002), Sopharak, A. and Uyyanonyara, B. (2006), Niemeijer, M. et al.
(2005), Wang, H. et al. (2000), Sinthanayothin, C. et al. (2002), Hipwell, J.H. et al. (2000), Li,
H. and Chutatape, O. (2000), Osareh, A. (2002), Ege, B.M. et al. (2000), Usher, D. et al.
(2004) and Abramoff, D.M. et al. (2008) focused on detecting selected lesions as will be
lbrther explained in Section 2.4.2. This does not provide useful data on the performance of
'IDtomatic detection of retinal abnormalities. The related work by Zhang, L. et al. (2009)
3
1
1.3
shows an improved detection of blood vessel which is able to include neovascularization net.
However, the work did not classify the neovascularization from normal blood vessel.
The clinical eye problem is however different from the experimental situation. Usually, a
patient will have more than one type of lesion. Thus, to focus on detecting a selected lesion is
not an appropriate approach if the detection method is to be applied clinically. This research
aims to simulate the clinical problem condition by having more than one lesion in the retinal
image. This approach is important as it will allow researchers to observe the effect of each
parameter set in the algorithm dynamically, because the number of lesions is not fixed as one.
Only a few researchers did the study on neovascularization that is the advanced
proliferative diabetic retinopathy condition. This condition is much severe than having only
dark lesions or bright lesions as it involves branches of blood vessels that could lead to
bleeding and blindness. This research includes an attempt to detect neovascularization in the
retinal image.
SCOPE AND OBJECTIVES
J
The focus of this research is to contribute a reliable image processing method for
automatic detection of diabetic retinal abnormalities. A limitation for this research is the
standardization of the database cannot be realized due to various specification of fundus
camera used. This research obtains the retinal images from several website and also from a
local hospital. Thus, the database contains different image resolution, angle of lens and quality.
4
Pusat Khidmat Maklumat Akadt~ UNIVERSm MALAYSIA SARAWAK
It affects and lower the performance of the proposed algorithm as the algorithm needs to cater
with the differences from various images.
The objectives of this research are:
i. Identify the progression of retinopathy in eyes and the symptoms.
11. To develop a method for automated detection of diabetic retin
fundus photographic with adaption to image quality.
iii. To diagnose and classify dark retinal, bright retinal and
abnormalities as an abnormal and normal retinal images
opathy in digitized
neovascularization
1.4 RESEARCH OVERVIEW
The overall structure of the methods used in this research is shown in Figure 1.1. The
input colour retinal image or fundus image is analyzed automatically and produces an
assessment of abnormalities characteristic. The proposed method will then be able to identify
and quantify the characteristic into particular classifications.
Detection of retinal abnormalities in this research is classified into microaneurysms and
haemorrhages as dark lesions, hard exudates and cotton wool spots as bright lesions and
aeovascularizations. However, detection and classification of neovascularization abnormality
bas not been explored widely compared to dark and bright lesions. Therefore, this research is
III attempt to explore the advanced proliferative diabetic retinopathy condition by using
digital image processing. The result of obvious detected retinal abnormalities will be used to
classify the nonnal or abnormal retinal images.
5
Input Image \
+ .. J I Bright lesions
f Dark lesions Neovascularization I
I
J
Figure 1.1 : Outline of the proposed system for automatic detection of retinal abnonnalities.
1.5 THESIS STRUCTURE
The thesis consists of five chapters and is organized as follows: Chapter 2 introduces the
Rtina structure and its complication, i.e. diabetic retinopathy background, symptoms, causes
treatment. This chapter also provides literature review on retinopathy techniques. Chapter
describes the retinal abnonnalities detection methodology such as image acquisition, pre
_~;sing. segmentation and classification of retinal abnonnalities with the proposed image
_~lSinlg method. Chapter 4 reports analysis of the proposed method in tenns of sensitivity
specificity together with graph of ROC (receiver operating characteristic) for the retinal
available in this research. Chapter 5 presents overall conclusion of this work, its
lllievemelnt and future investigations.
6
! Abnonnalities
detection
1 I
Classification
I
1 Abnonnal / Nonna!
CHAPTER 2
DETECTION OF RETINAL ABNORMALITIES USING IMAGE PROCESSING
2.1 INTRODUCTION
In this chapter, the structure of retina will be introduced especially the main retinal
components and retinal abnormalities. The prospective of diabetic retinopathy in terms of
disease progression, symptoms, causes and the treatments also will be discussed in this
section. Furthermore, previous works of automatic retinal image analysis are also reviewed in
this research as a guideline for developing the proposed method.
2.1 STRUCTURE OF THE EYE
Diabetic complication associated with the eye is called diabetic retinopathy disease.
Diabetic retinopathy occurs in the retina with the progression of abnormalities in the eye.
Figure 2.1(a) shows the cross sectional view of human eye and the image was taken from
National Eye Institute, 2008. The cornea is a see-through layer at the front of the eye and acts
like a window. Iris is responsible for the colour part of the eye. Pupil is located in the middle
of iris which controls the amount of light that enters the eye. Lens are located behind the
comea, transparent and shaped like a disc. Lens control the focus of the eye. Optic nerve is
responsible to deliver messages from the retina to the brain. Vitreous gel is a clear gel that
7
fills the space between the lens and the retina. The retina is a light sensitive part inside the
inner layer of the eye. It is responsible to turn light into image signal that the brain can
understand.
(a) (b)
Figure 2.1: Structure of eye: (a) Cross sectional view of human eye (b) A digital photograph of the retina or fundus
Image
For the research work presented in this thesis. the retina is the most important part of the
. The retina acts like a tilm as shewn in Figure 2.1 (b) and the image was taken from
J
lainen. M. 2005. Once the light reaches the retina, impulses are created and sent via
'c disc nerve to the brain. The retina colour is normally reddish due to the blood supply
retina. Optic disc. macula and blood vessel are the main anatomical structures in retina.
8