Decision Support System for Detection of Hypertensive Retinopathy Using AVR and Papilledema

Abstract

The interior and vital part of human eye is retina whose function is to capture and send images to brain. It consists of different structures along with two types of blood vessels; veins and arteries. These retinal blood vessels are affected by number of eye diseases such as Hypertensive Retinopathy (HR) and Diabetic Retinopathy (DR). HR is a retinal disease that is caused by consistent elevated blood pressure (hypertension). Many people in the World are suffering from HR disease; however, in most of cases, HR patients are unaware of it. The automated diagnostic systems are very useful for ophthalmologists to diagnose different retinal diseases. With the help of automated systems, the ophthalmologists can monitor and make treatment plan of retinal disease. Many researchers have developed different automated HR detection systems, but no automated system exists that detects and grades HR along with Papilledema (last stage of HR). Most of existing methods only performed artery venous classification rather than complete automated method for HR detection and grading. In this thesis, an automated system is presented that detects the HR at various stages using Arteriovenous Ratio (AVR) and Papilledema (optic disc swelling) signs. The proposed system consists of two modules i.e. vascular analysis for calculation of AVR and optic nerve head region analysis for Papilledema. AVR calculating stage consists of three major modules i.e. main component extraction, Artery and Vein (A/V) classification and AVR calculation. A new set of color and statistical features have been proposed in this research for accurate A/V classification. The proposed system effectively performs A/V classification and vessels width calculation for AVR computation to diagnose and grade HR. Second module detects and grades the Papilledema through analysis of fundus retinal images. The proposed system formulates a feature set which consists of Grey-Level Co-occurrence Matrix, optic disc margin obscuration, color and vascular features. A feature vector of these features is used for classification of normal and Papilledema images using Support Vector Machine (SVM) with its Radial Basis Function (RBF) kernel. The variations in retinal blood vessels, color properties, texture deviation of optic disc and its peripapillary region, and fluctuation of obscured disc margin are effectively identified and used by the proposed system for the detection and grading of Papilledema. In this thesis, a new local dataset AVRDB containing 100 images is developed for analysis of HR and annotated with assistance of expert ophthalmologists of Armed Forces Institute of Ophthalmology (AFIO), Pakistan. The proposed methods are evaluated on the images of INSPIRE-AVR, VICAVR, STARE and newly developed HR dataset (AVRDB). The proposed HR detection method shows the average accuracies of 95.14%, 96.82% and 98.76% for INSPIRE-AVR, VICAVR and AVRDB databases, respectively. It also shows HR grading results with average accuracies of 98.65%, 98.61% and 98.92% for INSPIRE-AVR, VICAVR and AVRDB databases, respectively. The proposed Papilledema detection method shows average accuracy of 92.86% and grading results with average accuracy of 97.85% on hybrid dataset of 160 images (70 images of AVRDB database and 90 images of STARE database), respectively. These results authenticate that this research is a milestone towards automated detection and grading of HR disease.