An Automated Deep Learning Based Clinical Decision Support System for Early Detection of Diabetic Retinopathy from Fundus Images

Abstract

Diabetic Retinopathy (DR) is one of the leading causes of preventable blindness among diabetic patients, making early detection and timely diagnosis essential for effective treatment and vision preservation. Manual examination of retinal fundus images is often time-consuming and depends on the expertise of ophthalmologists, creating the need for reliable automated screening systems. This study proposes an intelligent deep learning-based framework for the early detection and severity grading of Diabetic Retinopathy using retinal fundus images. The proposed approach employs a hybrid deep learning architecture that combines a U-Net encoder for accurate lesion segmentation with a ResNet50-based classifier for automated classification of retinal images into five clinically recognized DR severity levels. The segmentation module effectively identifies retinal abnormalities such as microaneurysms, hemorrhages, and exudates, while the classification module accurately predicts the stage of the disease. The proposed framework improves diagnostic performance by achieving over 7% higher classification accuracy compared with conventional CNN-based models and demonstrates robust performance across diverse retinal image datasets. By providing accurate, efficient, and automated screening, the proposed system assists ophthalmologists in early diagnosis, reduces diagnostic workload, and supports timely clinical intervention. Overall, the framework offers a scalable and reliable computer-aided diagnostic solution for Diabetic Retinopathy screening and contributes to improved eye healthcare through intelligent deep learning techniques.