Diagnosis and Grading of Diabetic Retinopathy using Deep Learning

Abstract

Diabetic retinopathy (DR), which causes tissue on the eye that damages visibility, is a common complication of type-2 diabetes. If it is not discovered in time, total blindness might occur. DR is irreversible. DR is primarily among adults who are of working age. More than 150 million people are affected by diabetic retinopathy (DR), which accounts for 2.6% of blindness worldwide. Different indications of DR are vision distortion, bulging of the eye, and formation of irregular blood vessels. The traditional way is to use Computer-aided Diagnosis (CAD) systems during treatment. The dataset used is the APTOS blindness detection dataset that is accessible in Kaggle. The Convolutional Neural Networks (CNN) is the most effective way for classifying images. In this paper, the MobileNet architecture, a deep learning technique is utilized to automate the diagnosis of the disease and estimate the severity of the eye into several stages through which the accuracy obtained for training is 95% and validation is 82%.