Detecting Diabetic Retinopathy Using Image Processing

Diabetic Retinopathy (DR) is a progressive eye disorder caused by diabetes that can severely impair vision, often developing silently in its initial stages without noticeable symptoms. Early and accurate detection of DR is crucial to initiate timely interventions and reduce the risk of permanent vision loss. This study introduces an automated diagnostic framework for DR classification utilizing a deep learning approach built on a Convolutional Neural Network (CNN). The system adopts the ResNet-50 architecture, trained on a carefully prepared dataset of retinal fundus images categorized into five clinical stages: No DR, Mild, Moderate, Severe, and Proliferative DR. To ensure balanced training, the dataset contained 100 samples for each class. The training process involved extracting features and learning hierarchical patterns through the ResNet-50 layers, enabling the model to recognize subtle indicators across various DR stages. Following the training, a dedicated testing module was implemented to analyze multiple input images concurrently and predict their respective DR severity levels. The model demonstrated reliable performance on unseen images, highlighting its potential for broader application. This framework offers an effective tool to improve DR screening processes, particularly in regions with limited access to specialized eye care. By supporting earlier detection and reducing the burden on healthcare professionals, the proposed system contributes to advancing AI-powered diagnostic solutions for diabetic retinopathy.