Artificial Intelligence for Real-Time Any-Stage Diabetic Retinopathy Detection: Systematic Review and Meta-analysis

Abstract

Introduction: Diabetic retinopathy (DR) is a leading cause of blindness among working-age adults worldwide. Traditional manual screening methods are often time-consuming and require specialized resources, limiting their accessibility. This review evaluates the diagnostic accuracy of artificial intelligence (AI)–assisted retinal imaging for early DR detection in patients with diabetes mellitus. Methods: A PRISMA-based systematic review was conducted across Google Scholar, ProQuest, PubMed, and ScienceDirect for studies published between 2015 and 2025. We assessed study quality and risk of bias using QUADAS-2 and ROBINS-I. Results: 21 studies were included, with 6 eligible for meta-analysis. For detection of any DR, pooled sensitivity was 0.92 (95% CI: 0.81–0.97) and pooled specificity was 0.90 (95% CI: 0.84–0.94). The pooled diagnostic odds ratio was 77.1 (95% CI: 40.1–148.3). The log diagnostic odds ratio (log DOR) was 4.35 (95% CI: 3.69–5.00), corresponding to a pooled DOR of 77.1 (95% CI: 40.1–148.3). The SROC curve demonstrated high diagnostic accuracy with moderate inter-study variability. IAQ showed superior diagnostic performance compared to SAQ. Conclusion: AI-assisted retinal imaging demonstrates high diagnostic accuracy for early DR detection. Future prospective studies focusing on DR severity staging are warranted to validate these findings further.