Detection of Retinal Lesions Based on Deep Learning for Diabetic Retinopathy

Abstract

Diabetic retinopathy (DR), is a physical condition that appear due to damages in the vessels of retina. It can occur if person have type one or type two diabetics. Also it occur due to high sugar levels in blood. At starting there is only mild vision problems eventually lose sight. It is an ordinary eye disease found in people with diabetes. This paper automatically and efficiently detect and classify the severity of DR. The first stepis Pre-processing, here perform Green channel extraction, Blood vessel extraction and Optic Disc (OD)removal. Green channel extraction is done to enhance the contrast. Kernel fuzzy c-means is usedto extract blood vessels and OD is removed by morphological operation. The next step isRecognition of Diabetic features, in this first is to recognize Hard Exudates, which is based on recursive region growing segmentation (RRGS) algorithm. The second one is recognition of Hemorrhages (HEM) and Micro aneurysms (MA) by using Matched Filtering, Laplacian of Gaussian Filtering, and Mutual Information Maximization using DE. From these extract features such as the microneurysms (MAs) counts, perimeter, area and exudate count, the area and perimeter of blood vessels. Then the extracted features are fed to CNN for classification purpose. This method reducing the workload of an ophthalmologist with an accuracy of around 98%.