Why Isn't AI in Eye Clinics Yet? A Systematic Review of Barriers and Pathways for AI-Based Fundus Image Diagnostics
Synopsis
This systematic review evaluates 34 studies from 2018 to 2025, finding that AI often exceeds 90% accuracy and can match or outperform expert clinicians in diagnosing common ocular diseases such as diabetic retinopathy, glaucoma, retinopathy of prematurity, and age-related macular degeneration, yet real-world deployment remains constrained by three gaps—disjointed integration into clinical workflows, lack of transparency in AI decision-making, and poor generalizability across diverse populations—and it proposes actionable pathways to bridge the "last-mile gap" between research and clinical practice.
Interpretation
The review synthesizes 34 studies (2018-2025) showing that AI often exceeds 90% accuracy in diagnosing common ocular diseases including DR, glaucoma, ROP, and AMD, in some cases matching or even outperforming expert clinicians. It consolidates diagnostic performance evidence scattered across different diseases and years into a single overall picture of current AI fundus image diagnostic capability. Based on a systematic review of 34 published studies; the performance figures come from those studies' reported results rather than a new single-center experiment.
Three significant gaps in real-world deployment are identified: disjointed integration into clinical workflows, lack of transparency in AI decision-making, and poor generalizability across diverse populations. It breaks down the phenomenon of high laboratory accuracy but limited clinical adoption into three separately addressable barriers, rather than attributing it vaguely to immature technology. Derived from comprehensive analysis of current literature, representing a synthesis of reported findings and discussions rather than newly collected primary data.
The review proposes actionable pathways to bridge the "last-mile gap" between research and clinical practice, aiming toward equitable, scalable, and trustworthy solutions for global vision care. It goes beyond listing barriers to offer an implementation-oriented framework, translating review conclusions into directions for advancing adoption. A framework-level recommendation based on literature synthesis, whose force depends on subsequent validation in real clinical settings.
Perspective
The review's conclusions are intended for researchers, clinical teams, and deployers seeking to bring AI-based fundus image diagnostics into real ophthalmic care and screening settings, applicable to diagnostic contexts for common ocular diseases such as DR, glaucoma, ROP, and AMD; the proposed pathway framework aims to guide translation from research to clinical practice rather than to replace local validation in specific settings.
Readers may still watch: how the relative weight of the three barriers varies across health systems and resource settings; the feasibility and validation approach of the proposed pathways in real clinical environments; and the differences among studies behind the phrase "often >90% accuracy" in terms of populations, devices, and evaluation methods. This summary is based on abstract-level text only, without the figures and study-by-study details of the original, so its characterization of differences among specific studies is limited.