Identifying of diabetic retinopathy from fundus image using deep learning

Abstract

Diabetic retinopathy is one of the complications of diabetes that affects the retina of the eyes resulting to blindness if not diagnosed in time. This is initiated by the destruction of the blood vessels in the layer of light-sensitive tissue located at the back part of the eye (retina). Finally, in this study, a deep learning-based solution is developed to identify and classify diabetic retinopathy from the fundus images accurately. The dataset used includes 5,200 fundus images categorized into five classes: It can be categorized as having No DR, Mild, Moderate, Severe, and Proliferative DR. Five types of CNN models were used for the study: Xception, VGG19, InceptionV3, MobileNetV2, and a newly developed CNN architecture. Thus, after evaluating, our own designed CNN model, we got maximum accuracy of 90.63% as compared to other models such as MobileneV2 (80.01%) and InceptiomV3 (76.62%). Based on the findings of the paper, our approach can be used to successfully identify diabetic retinopathy, which will go a long way in assisting and arriving at early diagnosis and treatment to alleviate the consequences of the illness, including blindness among patients.