Enhanced Detection of Diabetic Retinopathy Stages Using RetinaNet-DR: A Deep Learning-Based Approach

Diabetic Retinopathy (DR) is a leading cause of vision loss, particularly in individuals with prolonged diabetes. Timely detection and classification of DR stages are crucial for effective management and prevention of further complications. This paper introduces a novel automated classification model named RetinaNet-DR, specifically designed to categorize DR into five progressive stages: Normal, Mild, Moderate, Severe, and Proliferative DR, utilizing Optical Coherence Tomography (OCT) and Fundus images. The proposed system begins with a tailored preprocessing pipeline, enhancing image quality by applying normalization and noise reduction, followed by region-of-interest (ROI) extraction to ensure high-precision feature input. RetinaNet-DR incorporates a Convolutional Neural Network (CNN) with a ResNet-based architecture for accurate feature extraction, paired with an attention mechanism to highlight critical retinal features corresponding to DR severity. Experimental evaluations show that RetinaNet-DR achieves an impressive classification accuracy rate of 97.39% across all stages, surpassing conventional methods and exhibiting robust performance compared to recent models. This classification system has the potential to support clinicians by enabling more precise and early detection of DR stages, thereby improving patient care and facilitating timely intervention. Future developments will explore integrating multi-modal imaging sources to further enhance the model's clinical applicability and diagnostic performance.