Automated image detection of retinal pathology

Diabetes is approaching pandemic numbers, and as an associated complication, diabetic retinopathy is also on the rise. Much about the computer-based diagnosis of this intricate illness has been discovered and proven effective in research labs. But, unfortunately, many of these advances have subsequently failed during transition from the lab to the clinic. So what is the best way to diagnose and treat retinopathy? Automated Image Detection of Retinal Pathology discusses the epidemiology of the disease, proper screening protocols, algorithm development, image processing, and feature analysis applied to the retina.

Conveys the Need for Widely Implemented Risk-Reduction Programs

Offering an array of informative examples, this book analyzes the use of automated computer techniques, such as pattern recognition, in analyzing retinal images and detecting diabetic retinopathy and its progression as well as other retinal-based diseases. It also addresses the benefits and challenges of automated health care in the field of ophthalmology. The book then details the increasing practice of telemedicine screening and other advanced applications including arteriolar-venous ratio, which has been shown to be an early indicator of cardiovascular, diabetes, and cerebrovascular risk.

Although tremendous advances have been made in this complex field, there are still many questions that remain unanswered. This book is a valuable resource for researchers looking to take retinal pathology to that next level of discovery as well as for clinicians and primary health care professionals that aim to utilize automated diagnostics as part of their health care program.

Author Notes

Herbert Jelinek, Charles Stuart University, Albury, New South Wales, Australia

Michael J. Cree, University of Waikato, Hamilton, New Zealand

H. F. Jelinek and M. J. CreeD. Worsley and D. SimmonsM. D. Abrámoff and M. NiemeijerL B. BäcklundA. OsarehM. J. CreeN. Cheung and T. Y. Wong and L. HodgsonJ. V. B. Soares and R. M. Cesar Jr.K. H. Fritzsche and C. V. Stewart and B. RoysamN. W. Witt and M. E. Martínez-Pérez and K. H. Parker and S. A. McG. Thorn and A. D. HughesK. Yogesan and F. Reinholz and L J. Constable

Preface	p. xiii
Contributors	p. xvii
1 Introduction	p. 1
1.1 Why Automated Image Detection of Retinal Pathology?	p. 1
1.1.1 The general clinical need	p. 2
1.1.2 Diabetes: A global problem	p. 2
1.1.3 Diabetic retinopathy	p. 2
1.1.4 Eye-screening for diabetic retinopathy	p. 3
1.1.5 Other retinal pathologies	p. 5
1.1.6 The retina as an indicator for disease elsewhere	p. 6
1.1.7 Research needs in automated retinopathy detection	p. 6
1.1.8 The engineering opportunity	p. 7
1.2 Automated Assessment of Retinal Eye Disease	p. 7
1.2.1 Automated microaneurysm detection in diabetic retinopathy	p. 8
1.2.2 Hemorrhages	p. 9
1.2.3 White lesion segmentation	p. 9
1.2.4 Localization of important markers	p. 10
1.2.5 Retinal vessel diameter changes in disease	p. 11
1.2.6 Retinal blood vessel segmentation	p. 11
1.2.7 Mathematical analysis of vessel patterns	p. 12
1.3 The Contribution of This Book	p. 13
2 Diabetic Retinopathy and Public Health	p. 27
2.1 Introduction	p. 27
2.2 The Pandemic of Diabetes and Its Complications	p. 28
2.3 Retinal Structure and Function	p. 29
2.4 Definition and Description	p. 35
2.5 Classification of Diabetic Retinopathy	p. 40
2.6 Differential Diagnosis of Diabetic Retinopathy	p. 40
2.7 Systemic Associations of Diabetic Retinopathy	p. 42
2.7.1 Duration of diabetes	p. 42
2.7.2 Type of diabetes	p. 42
2.7.3 Blood glucose control	p. 42
2.7.4 Blood pressure	p. 42
2.7.5 Serum lipids	p. 43
2.7.6 Renal disease	p. 43
2.7.7 Anemia	p. 43
2.7.8 Pregnancy	p. 43
2.7.9 Smoking	p. 43
2.8 Pathogenesis	p. 43
2.8.1 Hyperglycemia	p. 43
2.8.2 Hematological abnormalities	p. 44
2.8.3 Leukostasis and inflammation	p. 44
2.8.4 Growth factors	p. 44
2.8.5 Neurodegeneration	p. 45
2.9 Treatment	p. 45
2.9.1 Management of systemic associations	p. 45
2.9.2 Ocular treatments	p. 45
2.9.3 Investigational treatments	p. 46
2.10 Screening	p. 48
2.10.1 Methods of screening	p. 48
2.10.2 Frequency of screening	p. 54
2.10.3 Cost effectiveness of screening	p. 54
2.10.4 Access to care and screening	p. 54
2.11 Conclusion	p. 55
3 Detecting Retinal Pathology Automatically with Special Emphasis on Diabetic Retinopathy	p. 67
3.1 Historical Aside	p. 67
3.2 Approaches to Computer (Aided) Diagnosis	p. 68
3.3 Detection of Diabetic Retinopathy Lesions	p. 70
3.4 Detection of Lesions and Segmentation of Retinal Anatomy	p. 71
3.5 Detection and Staging of Diabetic Retinopathy: Pixel to Patient	p. 71
3.6 Directions for Research	p. 72
4 Finding a Role for Computer-Aided Early Diagnosis of Diabetic Retinopathy	p. 79
4.1 Mass Examinations of Eyes in Diabetes	p. 79
4.1.1 Motive for accurate early diagnosis of retinopathy	p. 80
4.1.2 Definition of screening	p. 81
4.1.3 Practical importance of the concept of screening	p. 81
4.1.4 Coverage and timely re-examination	p. 81
4.2 Developing and Defending a Risk Reduction Program	p. 82
4.2.1 Explaining why retinopathy is suitable for screening	p. 82
4.2.2 Understanding reasons for possible criticism	p. 83
4.2.3 Fulfilling criteria for screening tests	p. 83
4.2.4 Setting quality assurance standards	p. 84
4.2.5 Training and assessment	p. 84
4.3 Assessing Accuracy of a Diagnostic Test	p. 84
4.3.1 Predictive value, estimation, power	p. 85
4.3.2 Receiver operating characteristic curve	p. 87
4.3.3 Area under curve	p. 89
4.3.4 Covariates	p. 90
4.4 Improving Detection of Diabetic Retinopathy	p. 90
4.4.1 Improving work environment	p. 91
4.4.2 Going digital	p. 91
4.4.3 Obtaining clear images	p. 91
4.4.4 Avoiding loss of information	p. 92
4.4.5 Viewing images	p. 92
4.4.6 Ensuring accurate grading	p. 93
4.4.7 Organizing for success	p. 93
4.5 Measuring Outcomes of Risk Reduction Programs	p. 93
4.5.1 Reducing new blindness and visual impairment	p. 94
4.5.2 Counting people who lost vision	p. 94
4.5.3 Understanding the importance of visual impairment	p. 95
4.6 User Experiences of Computer-Aided Diagnosis	p. 96
4.6.1 Perceived accuracy of lesion detection	p. 97
4.6.2 Finding and reading evaluations of software for retinopathy diagnosis	p. 101
4.6.3 Opportunities and challenges for programmers	p. 102
4.7 Planning a Study to Evaluate Accuracy	p. 103
4.7.1 Getting help from a statistician	p. 103
4.7.2 Choosing a measurement scale	p. 103
4.7.3 Optimizing design	p. 104
4.7.4 Carrying out different phases of research	p. 108
4.7.5 An example from another field	p. 109
4.8 Conclusion	p. 110
4.9 Appendix: Measures of Binary Test Performance	p. 120
5 Retinal Markers for Early Detection of Eye Disease	p. 121
5.1 Abstract	p. 121
5.2 Introduction	p. 122
5.3 Nonproliferative Diabetic Retinopathy	p. 123
5.4 Chapter Overview	p. 124
5.5 Related Works on Identification of Retinal Exudates and the Optic Disc	p. 128
5.5.1 Exudate identification and classification	p. 128
5.5.2 Optic disc detection	p. 130
5.6 Preprocessing	p. 132
5.7 Pixel-Level Exudate Recognition	p. 134
5.8 Application of Pixel-Level Exudate Recognition on the Whole Retinal Image	p. 137
5.9 Locating the Optic Disc in Retinal Images	p. 139
5.9.1 Template matching	p. 141
5.9.2 Color morphology preprocessing	p. 141
5.9.3 Accurate localization of the optic disc-based snakes	p. 144
5.9.4 Optic disc localization results	p. 146
5.10 Conclusion	p. 148
6 Automated Microaneurysm Detection for Screening	p. 155
6.1 Characteristics of Microaneurysms and Dot-Hemorrhages	p. 155
6.2 History of Automated Microaneurysm Detection	p. 156
6.2.1 Early morphological approaches	p. 156
6.2.2 The "standard approach" to automated microaneurysm detection	p. 157
6.2.3 Extensions of the standard approach	p. 159
6.2.4 Other approaches	p. 162
6.2.5 General red lesion detection	p. 164
6.3 Microaneurysm Detection in Color Retinal Images	p. 165
6.4 The Waikato Automated Microaneurysm Detector	p. 167
6.4.1 Further comments on the use of color	p. 171
6.5 Issues for Microaneurysm Detection	p. 172
6.5.1 Image quality assessment	p. 172
6.5.2 Image compression implications	p. 173
6.5.3 Optic disc detection	p. 175
6.5.4 Meaningful comparisons of implementations	p. 175
6.6 Research Application of Microaneurysm Detection	p. 177
6.7 Conclusion	p. 178
7 Retinal Vascular Changes as Btomarkers of Systemic Cardiovascular Diseases	p. 185
7.1 Introduction	p. 185
7.2 Early Description of Retinal Vascular Changes	p. 186
7.3 Retinal Vascular Imaging	p. 187
7.3.1 Assessment of retinal vascular signs from retinal photographs	p. 187
7.3.2 Limitations in current retinal vascular imaging techniques	p. 187
7.4 Retina] Vascular Changes and Cardiovascular Disease	p. 189
7.4.1 Hypertension	p. 189
7.4.2 Stroke and cerebrovascular disease	p. 191
7.4.3 Coronary heart disease and congestive heart failure	p. 193
7.5 Retinal Vascular Changes and Metabolic Diseases	p. 194
7.5.1 Diabetes mellitus	p. 196
7.5.2 The metabolic syndrome	p. 196
7.5.3 Overweight and obesity	p. 197
7.6 Retinal Vascular Changes and Other Systemic Diseases	p. 197
7.6.1 Renal disease	p. 197
7.6.2 Atherosclerosis	p. 198
7.6.3 Inflammation and endothelial dysfunction	p. 198
7.6.4 Subclinical cardiac morphology	p. 200
7.7 Genetic Associations of Retinal Vascular Changes	p. 200
7.8 Conclusion	p. 201
7.A Appendix: Retinal Vessel Caliber Grading Protocol	p. 201
7.A.1 Grading an image	p. 202
7.A.2 Example of the grading process	p. 204
7.A.3 Obtaining results	p. 205
7.A.4 Saving data	p. 206
8 Segmentation of Retinal Vasculature Using Wavelets and Supervised Classification: Theory and Implementation	p. 221
8.1 Introduction	p. 221
8.2 Theoretical Background	p. 224
8.2.1 The 1-D CWT	p. 224
8.2.2 The 2-D CWT	p. 225
8.2.3 The 2-D Gabor wavelet	p. 228
8.2.4 Supervised classification	p. 229
8.2.5 Bayesian decision theory	p. 231
8.2.6 Bayesian Gaussian mixture model classifier	p. 231
8.2.7 K-nearest neighbor classifier	p. 233
8.2.8 Linear minimum squared error classifier	p. 234
8.3 Segmentation Using the 2-D Gabor Wavelet and Supervised Classification	p. 235
8.3.1 Preprocessing	p. 235
8.3.2 2-D Gabor wavelet features	p. 237
8.3.3 Feature normalization	p. 238
8.3.4 Supervised pixel classification	p. 239
8.3.5 Public image databases	p. 240
8.3.6 Experiments and settings	p. 241
8.3.7 ROC analysis	p. 242
8.4 Implementation and Graphical User Interface	p. 245
8.4.1 Overview	p. 245
8.4.2 Installation	p. 246
8.4.3 Command line interface	p. 246
8.4.4 Graphical user interface	p. 247
8.5 Experimental Results	p. 249
8.6 Conclusion	p. 258
8.6.1 Summary	p. 258
8.6.2 Future work	p. 258
9 Determining Retinal Vessel Widths and Detection of Width Changes	p. 269
9.1 Identifying Blood Vessels	p. 270
9.2 Vessel Models	p. 270
9.3 Vessel Extraction Methods	p. 271
9.4 Can's Vessel Extraction Algorithm	p. 271
9.4.1 Improving Can's algorithm	p. 272
9.4.2 Limitations of the modified Can algorithm	p. 275
9.5 Measuring Vessel Width	p. 276
9.6 Precise Boundary Detection	p. 278
9.7 Continuous Vessel Models with Spline-Based Ribbons	p. 279
9.7.1 Spline representation of vessels	p. 279
9.7.2 B-spline ribbons	p. 284
9.8 Estimation of Vessel Boundaries Using Snakes	p. 288
9.8.1 Snakes	p. 288
9.8.2 Ribbon snakes	p. 289
9.8.3 B-spline ribbon snake	p. 289
9.8.4 Cross section-based B-spline snakes	p. 292
9.8.5 B-spline ribbon snakes comparison	p. 293
9.9 Vessel Width Change Detection	p. 294
9.9.1 Methodology	p. 294
9.9.2 Change detection via hypothesis test	p. 296
9.9.3 Summary	p. 298
9.10 Conclusion	p. 298
10 Geometrical and Topological Analysis of Vascular Branches from Fundus Retinal Images	p. 305
10.1 Introduction	p. 305
10.2 Geometry of Vessel Segments and Bifurcations	p. 306
10.2.1 Arterial to venous diameter ratio	p. 306
10.2.2 Bifurcation geometry	p. 308
10.2.3 Vessel length to diameter ratios	p. 311
10.2.4 Tortuosity	p. 312
10.3 Vessel Diameter Measurements from Retinal Images	p. 312
10.3.1 The half-height method	p. 313
10.3.2 Double Gaussian fitting	p. 314
10.3.3 The sliding linear regression filter (SLRF)	p. 314
10.4 Clinical Findings from Retinal Vascular Geometry	p. 315
10.5 Topology of the Vascular Tree	p. 318
10.5.1 Strahler branching ratio	p. 321
10.5.2 Path length	p. 321
10.5.3 Number of edges	p. 321
10.5.4 Tree asymmetry index	p. 322
10.6 Automated Segmentation and Analysis of Retinal Fundus Images	p. 323
10.6.1 Feature extraction	p. 324
10.6.2 Region growing	p. 326
10.6.3 Analysis of binary images	p. 327
10.7 Clinical Findings from Retinal Vascular Topology	p. 328
10.8 Conclusion	p. 329
11 Tele-Diabetic Retinopathy Screening and Image-Based Clinical Decision Support	p. 339
11.1 Introduction	p. 339
11.2 Telemedicine	p. 339
11.2.1 Image capture	p. 340
11.2.2 Image resolution	p. 341
11.2.3 Image transmission	p. 342
11.2.4 Image compression	p. 342
11.3 Telemedicine Screening for Diabetic Retinopathy	p. 344
11.4 Image-Based Clinical Decision Support Systems	p. 346
11.5 Conclusion	p. 347
Index	p. 351

Available:*

On Order

Summary

Summary

Author Notes

Table of Contents