A Deep Learning Technique for Bi-Fold Grading of an Eye Disorder DR-Diabetic Retinopathy

Abstract

Nowadays, with increasing cases of diabetes, one should control the blood sugar as well as perform regular examination of eyes to prevent oneself from blindness. Any person having diabetes is likely to develop diabetic retinopathy (DR). DR is triggered by high blood sugar due to diabetes. After some time, having excessive amount of sugar in blood, can damage retina. When sugar jams the tiny blood vessels the eyes are damaged and this will affect the blood vessels and result in leakage of fluid. Millions of working aged adults suffers from loss of sight due to diabetic retinopathy. DR cannot be treated completely, but early detection of DR prevents the person from vision loss. We proposed a deep learning model for detection of diabetic retinopathy. Detection of DR is a slow process. Physical detection of DR involves a trained clinician to study and estimate the color fundus photographs of the retina. Normal process of identification takes a minimum of two days. In our paper, convolutional neural network architecture has been used to classify images into two classes which is no-diabetic retinopathy and with diabetic retinopathy. APTOS-2019 blindness detection dataset has been used from Kaggle which contains high-resolution retinal images. Those images are used to train the model. Web-based interface has been created for easy interaction with the model.