Background Diabetic retinopathy (DR) remains a leading cause of preventable visual impairment among individuals with diabetes mellitus.[1,2] Screening programs at secondary care hospitals are often constrained by limited specialist availability. Artificial intelligence (AI)–based systems may provide an effective adjunct for early detection of DR.[3-7] Objective To evaluate the diagnostic performance of an AI-based system for detecting diabetic retinopathy using color fundus photographs in a secondary care hospital setting. Methods This retrospective observational study included fundus photographs of 100 patients with diabetes mellitus. Images were independently evaluated by an AI-based DR detection system and by a retina specialist. Specialist grading according to the International Clinical Diabetic Retinopathy Severity Scale[8] served as the reference standard. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and inter-observer agreement (Cohen’s kappa) were calculated. Results The AI system demonstrated high sensitivity and specificity for detecting any diabetic retinopathy. Substantial agreement was observed between AI output and specialist grading (Cohen’s κ = 0.79). Conclusion AI-based screening systems demonstrate promising diagnostic performance for diabetic retinopathy detection in secondary care hospital settings and may help address manpower shortages in resource-constrained environments.

Diabetic Retinopathy, Screening, Artificial Intelligence (AI).