Deep Learning Framework for Early Detection and Risk Analysis of Diabetic Retinopathy

Abstract

Retinopathy is a common and possibly sight-threatening disease that needs to be diagnosed correctly and quickly. This study examines how deep learning models, like ResNet-50 and a Simple CNN, can detect retinopathy in retina images. We use a dataset of Kaggle Diabetic Retinopathy Detection Training Dataset (DRD) retinal images, process them so they are all the same size, and train the models on a deep learning system at the cutting edge. We aim to compare how well these models work on a test dataset regarding accuracy. The results show that ResNet-50 does better than the Simple CNN baseline in all evaluation measures. It is more accurate and knows more about the features of retinopathy. The model is 94.6% accurate, which shows that it could be used in clinical settings. However, neither model can be generalized, especially since it needs to be tested on more patients from different groups.