AI-Powered Early Detection of Diabetic Retinopathy: A Deep Learning Approach for Improved Clinical Decision-Making

Abstract

Diabetic retinopathy (DR) is one of the leading causes of preventable blindness worldwide, and its early detection remains critical for effective treatment and clinical decision-making. This study presents an AI-powered deep learning framework for automated identification of diabetic retinopathy from retinal fundus images. A convolutional neural network (CNN) architecture, fine-tuned using transfer learning on the Kaggle EyePACS dataset, was employed to classify retinal images into five severity levels no DR, mild, moderate, severe, and proliferative DR. The proposed model integrates data augmentation, contrast enhancement, and adaptive learning rate optimization to improve detection robustness and interpretability. The system’s performance was evaluated using accuracy, sensitivity, specificity, and area under the ROC curve (AUC), achieving an overall accuracy of 96.2% and an AUC of 0.982 on the test dataset. Additionally, Grad-CAM visualization was applied to highlight lesion regions, enhancing the model’s explainability for clinical practitioners. The results demonstrate that the proposed deep learning approach not only achieves superior classification performance but also provides a reliable, interpretable decision-support tool for ophthalmologists. This study highlights the transformative potential of artificial intelligence in ophthalmic diagnostics, supporting faster, objective, and more accurate screening of diabetic retinopathy in clinical settings.