The Retina’s Warning Signal: Artificial Intelligence Prediction of Diabetic Retinopathy Progression

Abla Almalik¹, Asim Ahmed², Antoine Lefevre³

Abstract

Diabetic retinopathy (DR) remains a major cause of preventable blindness worldwide. Early identification of patients at risk of rapid disease progression is essential for preventing severe visual impairment. Artificial intelligence (AI) has recently shown promise in analyzing retinal images and predicting disease outcomes. This study evaluated the performance of a deep-learning model in predicting diabetic retinopathy progression using baseline fundus photographs.

A retrospective cohort study was conducted using 21,470 retinal fundus images obtained from 7,960 patients with diabetes between 2020 and 2024. The AI model was trained to predict progression to vision-threatening diabetic retinopathy (VTDR) over a three-year follow-up period.

During the study period, 15.2% of patients showed disease progression, while 6.4% developed vision-threatening diabetic retinopathy. The deep learning model achieved an area under the curve (AUC) of 0.93, with sensitivity of 88.6% and specificity of 85.9% for predicting VTDR.

Risk stratification analysis revealed that patients in the highest AI-predicted risk group accounted for 43% of future severe cases, despite representing only 18% of the total cohort. Simulation models indicated that implementing AI-based risk stratification could reduce routine screening visits by 24% while maintaining detection of over 92% of severe cases.

These findings highlight the potential of artificial intelligence to transform diabetic retinopathy screening programs by enabling personalized follow-up intervals and early identification of patients at high risk of vision-threatening disease.