Retinal imaging and artificial intelligence: A systematic review and meta-analysis of diagnostic techniques for
Zohreh Ganji1, Farzaneh Nikparast1, Naser Shoeibi2
1Department of Medical Physics, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran; Student research committee, Mashhad University of medical sciences, Mashhad, Iran.
Background:
Artificial intelligence (AI)-driven analysis of retinal images holds promise for noninvasive, early detection of neurodegenerative disorders. We conducted a prospectively registered systematic review and meta‑analysis to quantify diagnostic accuracy of AI‑assisted retinal imaging across Alzheimer's disease (AD), Parkinson's disease (PD), and related conditions.
Methods:
We searched six databases for studies (2010-2024) employing AI/ML models for binary classification of neurodegenerative disease versus healthy controls. Risk of bias was assessed using QUADAS-2. We extracted pooled area under the receiver operating characteristic curve (AUC) with 95 % confidence intervals (CIs), heterogeneity statistics, and participant numbers. Certainty of evidence was graded using GRADE criteria.
Results:
Our analysis included ten studies (seven on AD, four on PD; 496 patients, 441 patients, and 36,990 healthy controls). The overall pooled AUC was 0.73 (95 % CI, 0.69-0.77; I²=78 %). Subgroup analyses showed an AUC of 0.72 for AD and 0.70 for PD. QUADAS-2 assessments indicated low-to-moderate risk of bias, with patient selection variability being a key concern. GRADE evaluations showed moderate to high certainty in the evidence.
Conclusions:
AI‑assisted retinal imaging demonstrates consistent, moderate accuracy for early detection of AD and PD. Future research should standardize imaging protocols and patient selection criteria, and pursue large-scale, prospective validation.
