Classification of Diabetic Retinopathy and Segmentation of Hard Exudates from Retinal Fundus Images

Abstract

Diabetic Retinopathy (DR) is a disorder of the retinal vasculature. It happens in nearly all patients with long-standing diabetes mellitus and can result in blindness. Screening of DR is essential for both early detection and early treatment of the disease. The presence of exudates is one of the complications of diabetes mellitus that is considered as the major cause of vision loss among people around the world. It results from leakage of fluid rich in fat and cholesterol from damaged retinal vasculature. The current trend for hard exudates (HE) detection is a visual grading method that is time-consuming and susceptible to observer errors. The computer-aided detection of HE would potentially assist in achieving a fast and accurate diagnosis. Even though numerous researches are done previously to come up with a method to detect DR, further improvement is needed for butter accuracy. This thesis study develops an automatic method for DR detection and subsequently develops an effective system for the segmentation of exudates. The segmented exudates indicate the presence of DR. Before the detection of exudates, the color retinal images which were collected from international databases IDRiD, Kaggle and e-optha Ex datasets and local images from JUMC were classified as images with DR or healthy using local binary pattern (LBP) texture descriptor and gray level co-occurrence matrix (GLCM) followed by segmentation of exudates using the morphological top-hat. Different performance metrics were used to measure the performance of the developed system. The system offered an accuracy of 98.8% for the classification of DR positive and healthy images. It offered an accuracy of 97.8%, sensitivity of 96.3%, specificity of 90%, dice score 70%, and jaccard similarity 90.9% for segmentation of exudates. The developed system also counts the number of exudates by counting white pixels in the image in each stage of the DR to just indicate how the number of exudates increases with the stage. The method was better in accuracy compared to other methods reported in the literature.