Artificial Intelligence in Eye Disease: A Systematic Review of Screening, Diagnosis, Prognosis, Treatment-Response Prediction, and Clinical Decision Support

Abstract

Background/Objectives: Artificial intelligence (AI) is expanding eye care beyond image-based detection toward prognosis, treatment-response prediction, monitoring, triage, and clinician-facing decision support. We evaluated AI across eye-disease screening, diagnosis, prognosis, treatment response, and decision support, emphasizing validation, explainability, and clinical readiness. Methods: This PROSPERO-registered review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement. Six databases plus supplementary routes identified English-language literature published from 1 January 2015 through 10 August 2026. Included studies underwent design-mapped Joanna Briggs Institute (JBI)/adapted-JBI item-level appraisal; heterogeneity precluded meta-analysis. Results: The evidence base comprised 270 studies. Diabetic retinopathy/diabetic macular oedema was the largest clinical area (82), followed by retinal/multi-disease AI (52), glaucoma (46), and age-related macular degeneration (AMD)/macular disease (29); convolutional neural network (CNN)/convolutional deep learning was the largest primary method group (136). External validation was confirmed in 130 studies, including 95 with a clinical/geographic/prospective/real-world external-validation component (86 clinical-only and nine mixed clinical + technical); 49 had a strict direct clinical comparator and 90 used explainability. Pediatric/pediatric-relevant evidence accounted for 30 studies, while explicit optometry/optometrist involvement appeared in only three. Conclusions: Ophthalmic AI spans diagnosis through clinical management, but translation remains limited by heterogeneous outcomes, incomplete independent validation, few direct comparators, and sparse evaluation of safety, workflow effects, and patient outcomes.