A Comprehensive Review of Deep Learning Approaches for Diabetic Retinopathy Detection
Km Meenakshi, Shakeeluddin, Abhishek Kumar, Deepshikha Bhargava
Abstract
Diabetic retinopathy is the primary reason of having diabetes and one of the leading causes of blindness in the world fall under retinal based disease. It is hybrid deep learning modeling that could become a game-changer in the early identification and categorization of Diabetic Retinopathy, with better patient outcomes and relieving health systems of pressure. The best of what deep learning architectures, specifically recurrent neural networks and convolutional neural networks, have to offer will be put together to use information emanating from retinal scans and patient data to raise the bar of the diagnostic system. It also provides an overview of the current databases, screening programs, performance indicators, and biomarkers involved in diabetic retinopathy. This survey emphasizes the accuracy of the different hybrid models on different datasets and highest accuracy of 98.60% was attained on the APTOS, MESSIDOR2, and IRDid datasets using the RetNet-10 model and VGG16 has very low accuracy at 71.65% on EyePACS dataset.