Effectiveness of screening modalities for early detection of diabetic retinopathy: a systematic review and meta-analysis of tele-ophthalmology, AI-based tools, and conventional methods

Background Diabetic retinopathy (DR) causes severe vision impairment that requires early screening methods for effective detection. The combination of Artificial Intelligence (AI) and tele-ophthalmology technology provides an effective solution that enhances both DR detection rates and patient access to care. Objective The study aims to assess how well AI-based telemedicine smartphone conventional and population-based screening methods detect diabetic retinopathy in terms of effectiveness, accuracy, and real-world performance. Methodology Researchers executed a comprehensive literature search across PubMed, Scopus, Web of Science, and Embase to find articles published between 2012 and 2025. The study examined three types of studies: AI-based DR screening, tele-ophthalmology, and evaluations of standard diagnostic methods. The analysis used random-effects meta-analysis to estimate pooled odds ratios (ORs), assess heterogeneity using I 2 , and test for publication bias with funnel plots and Egger’s test. Results The study included 45 different research studies. The standalone deep learning/AI tools (6 studies) demonstrated a pooled OR of 5.79 (95% CI: 5.22–6.42; p < 0.05), with low heterogeneity and no evidence of publication bias. The automated AI systems processed human grading data from 13 studies, yielding an OR of 5.48 (95% CI: 5.09–5.90; p < 0.05), indicating consistent effect sizes. The smartphone-based AI screening system showed an OR of 4.73 (95% CI: 3.96–5.66; p < 0.05), indicating moderate heterogeneity (I 2 = 54%). The tele-ophthalmology/remote screening system produced 11 studies reporting an OR of 4.91, whereas conventional physician screening produced 3 studies reporting an OR of 4.96, yielding consistent results with low heterogeneity. The population-based/community screening system produced 5 studies that demonstrated an OR of 4.90 (95% CI: 4.33–5.54; p < 0.05) and exhibited some signs of publication bias according to Egger’s test ( p = 0.019). All methods achieved statistically significant progress in DR detection research. Conclusion AI-based screening, including deep learning algorithms, automated grading, and tele-ophthalmology, shows high diagnostic accuracy and consistent effectiveness. The implementation of smartphone- and population-based approaches is feasible, yet their outcomes vary, necessitating validation through context-specific assessments.