Leveraging CNNs for Early Detection and Staging of Diabetic Retinopathy

Diabetic Retinopathy (DR) is a major cause of blindness, and its early detection is crucial hence the development of a reliable system for its detection. This work describes an automated DR detection system through Convolutional Neural Networks (CNNs) for image classification of retinal fundus photographs. To address key problems of how the model can deal with data variability, multi-class classification, and feature localization, our model is trained on a diverse set of data. CNN based method produced an accuracy of 93% and precision of 93.84 while recall was 95.57%. The studies presented in this paper show the system's possibility of DR screening in real time for a large population, which would ultimately improve DR detection and treatment. The further improvements are extending the database for generalization purposes and implementing the model into a convenient real-life tool for clinical usage.