Automatic Diagnosis of Diabetic Retinopathy Using Morphological Operations

Abstract

Diabetic retinopathy is diabetic eye disease or a sight threatening complication (one of the major cause of blindness) for the person suffering from diabetes which causes progressive loss to the retina, in which retina of the eye is affected because the capillaries of the retina are damaged. Diabetic Retinopathy is unpredictable at early stage, it is only predictable in advanced stage when diabetic patient suffers from loss of vision due to leakage of lipid, blood vessels bursts and there is formation of new fragile blood vessels which blocks the blood supply to retina. Diabetic Retinopathy include Microaneurysm, hemorrhage and exudates. However, early detection and treatment is most important that can reduce the chances of occurrences of blindness about 95%. To analyze Microaneurysm and hemorrhage as early stages of DR is a challenging task for Ophthalmologists to prevent vision loss. Automatic analysis of Diabetic Retinopathy helps in preventing vision loss. Our proposed method is based on automatic detection of hemorrhage using colorful fundus images. In proposed work we have used supervised learning to classify the data as hemorrhage and without hemorrhage with SVM classifier. To find hemorrhage and its severity, we have extracted statistical features (including standard deviation, energy, entropy and contrast of an image), used classification approach and then segmentation methods. After feature detection, Morphological Operations are applied to detect blood vessels and hemorrhage detection with help of segmentation technique. Here the threshold optimization, Grey Wolf Optimization (GWO) techniques are used in our proposed work for getting maximum accuracy, sensitivity and specificity performance metrics.