Deep Learning Framework for Predicting Diabetic Retinopathy Progression Using Longitudinal Retinal Imaging and Electronic Health Records

Diabetic retinopathy (DR) is one of the leading causes of vision impairment among individuals with diabetes, making early prediction of disease progression essential for timely intervention and improved clinical outcomes. Traditional diagnostic approaches primarily rely on single-time retinal images and often overlook longitudinal disease evolution and valuable clinical information available in electronic health records (EHRs). This paper proposes a deep learning framework that integrates longitudinal retinal fundus images with patient EHR data to accurately predict diabetic retinopathy progression. The framework employs a convolutional neural network for extracting spatial features from retinal images and a temporal learning module to capture disease progression across multiple clinical visits. Clinical attributes, including age, duration of diabetes, HbA1c, blood pressure, and other risk factors, are encoded and fused with imaging features using an attention-based multimodal fusion mechanism. The integrated representation enables robust progression prediction while improving interpretability and predictive performance. Experimental evaluation demonstrates superior accuracy, sensitivity, specificity, F1-score, and AUC compared with existing state-of-the-art methods. The proposed framework offers an effective clinical decision support tool for early risk assessment, personalized treatment planning, and prevention of vision loss in diabetic patients.