AUTOMATIC DETECTION OF DIABETIC RETINOPATHY USING CNN

Abstract

Diabetic Retinopathy (DR) is a leading cause of preventable blindness worldwide, primarily affecting individuals with long-term diabetes mellitus. Progressive damage from chronic hyperglycemia leads to microaneurysms, hemorrhages, exudates, macular edema, and vision impairment without early detection. Manual diagnosis through fundus examination is time-consuming, subjective, and inaccessible in rural healthcare environments. This research proposes a Convolutional Neural Network (CNN)-based framework for automated DR detection and classification using retinal fundus images. The model leverages annotated retinal image datasets to learn hierarchical features including microaneurysms, exudates, neovascularization, and hemorrhages. Preprocessing techniques—image normalization, resizing, contrast enhancement, and augmentation—improve robustness and generalization. The system classifies DR into five severity stages: No DR, Mild, Moderate, Severe, and Proliferative DR. Experimental evaluation demonstrates that deep learning-based detection achieves significantly improved accuracy, sensitivity, and specificity compared to traditional machine learning methods. The automated approach reduces diagnostic time and supports large-scale screening programs, contributing to reduced blindness rates and improved patient outcomes