A Deep Learning Framework for Detection and Classification of Diabetic Retinopathy in Fundus Images Using Residual Neural Networks

Diabetic retinopathy is a damage to the blood vessels in the eye caused by diabetes. Failure to diagnose it in time can result in permanent vision impairment or blindness. Regular disease screening has traditionally been a time-consuming and costly endeavour. Utilising computer technology to detect these diseases automatically would be an effective method. In recent years, we have seen a rise in interest in the use of Convolutional Neural Networks (CNNs), a subcategory of deep learning systems, as a potential method for the evaluation of medical images. To classify fundus images into different severity levels using the ResNet-50-CNN architecture is the primary focus of this work. The effectiveness of the algorithm can be evaluated using the Kaggle blindness dataset APTOS 2019. The paper will focus on both binary and multiclass classifications of DR. On the APTOS dataset, the accuracy for training and validating two-class models is 96% and 95%, respectively. The accuracy of training and validation for classification of multiple classes in the extended APTOS dataset is SO% and 77%, respectively.