Abstract

Detection and Classification of Diabetic Retinopathy Using Deep Learning present an automated diagnostic framework designed to combat one of the primary causes of global preventable blindness. Diabetic Retinopathy (DR) requires early intervention to prevent irreversible vision loss; however, manual screening remains labor-intensive and subject to inter-observer variability. This project addresses these challenges by developing a high-performance Convolutional Neural Network (CNN) architecture capable of classifying retinal fundus images into five distinct severity levels: No DR, Mild, Moderate, Severe, and Proliferative DR. Utilizing the APTOS dataset, the proposed system integrates advanced image preprocessing such as CLAHE (Contrast Limited Adaptive Histogram Equalization) and Gaussian blurring to standardize variations in lighting and image quality. The deep learning model is specifically engineered to identify and extract discriminative pathological features, including microaneurysms, hemorrhages, exudates, and neovascularization. To enhance generalization and mitigate overfitting, rigorous data augmentation strategies are employed during the training phase. The framework offers a scalable, objective, and efficient tool for clinical decision support. By automating the detection of subtle retinal abnormalities, the system significantly reduces the diagnostic burden on ophthalmologists and enables rapid screening in underserved regions. Ultimately, this AI-driven approach provides a robust solution for early DR diagnosis, facilitating timely treatment and improving long-term patient outcomes in the management of diabetic eye disease.