Intelligent System for Early Detection of Diabetic Retinopathy using Deep Learning

To analyse retinal fundus images for early diabetic retinopathy (DR) detection, this research introduces a deep learning-based classification and grading system. The model aims to automatically predict the presence and severity of DR, a progressive eye disease that can lead to vision loss in diabetic patients. Using high-resolution fundus images, the system applies a hybrid deep learning architecture—combining a U-Net encoder for lesion segmentation and a ResNet50-based classifier for categorization into five DR grades. This intelligent system enhances clinical decision-making by highlighting disease-specific retinal abnormalities and providing accurate diagnostic predictions. The model demonstrates an improvement of over 7% in accuracy compared to conventional CNN-based models and performs reliably across varied datasets.