AI-Powered Retinal Disease Detection System for Diabetic Retinopathy Screening

Abstract

A retinal condition linked to diabetes, diabetic retinopathy is one of the main causes of blindness in the elderly population. Early identification using blood vessel analysis is essential for avoiding vision loss since it is generally asymptomatic in its early stages and is characterized by progressive damage to retinal blood vessels that may leak or burst, impairing vision. An intelligent deep learning framework based on Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models is developed and compared with conventional machine learning and deep learning techniques like SVM, XGBoost, ResNet50, and DenseNet201 in this study, which proposes an automatic diabetic retinopathy detection system using color fundus images. To enhance the image and extract features, preprocessing methods such cropping, denoising, histogram equalization, scaling, data augmentation, and normalization are used to the APTOS 2019 dataset. According to experimental data, the GRU model performs competitively with 98.36% accuracy, 97.62% precision, 99.19% recall, and 98.40% F1-score, while the suggested LSTM model attains the maximum performance with 98.64% accuracy (ACC), 98.14% precision (PRE), 99.19% recall (REC), and 98.66% F1-score (F1). The proposed sequential deep learning models outperform the baseline models in terms of classification ACC and reliability for automated detection of DR and early clinical screening support.