Retinal imaging and artificial intelligence: A systematic review and meta-analysis of diagnostic techniques for neurodegenerative diseases.
Level 3 - non-randomized controlled study
Systematic review and meta-analysis of diagnostic observational and case-control studies
PubMed 40886956 · doi:10.1016/j.pdpdt.2025.104788
What was done
Prospectively registered systematic review and meta-analysis of six databases (2010–2024) evaluating artificial intelligence/machine learning models for binary classification of Alzheimer's disease (AD) and Parkinson's disease (PD) versus healthy controls via retinal imaging. Study quality was evaluated using QUADAS-2, pooled area under the ROC curve (AUC) with 95% confidence intervals (CIs) was calculated, and certainty was assessed with GRADE criteria.
What was found
Ten studies were analyzed (seven on AD, four on PD), comprising 496 AD cases, 441 PD cases, and 36,990 healthy controls. The overall pooled AUC was 0.73 (95% CI, 0.69–0.77) with substantial heterogeneity (I² = 78%). Subgroup analyses yielded an AUC of 0.72 for AD and 0.70 for PD. QUADAS-2 indicated low-to-moderate risk of bias, driven mainly by patient selection variability.
Why it matters
Retinal imaging provides a potential noninvasive biomarker modality, but current AI models demonstrate only moderate accuracy when discriminating neurodegenerative cases from healthy controls.
Limits
The review included only 10 studies with notable statistical heterogeneity (I² = 78%) and heavy case-control imbalance (937 cases vs. 36,990 controls). Binary classification against healthy controls does not assess performance in real-world clinical differential diagnosis. Sensitivity, specificity, and exact imaging modalities were not reported in the abstract.
Cited by
- supports AI applied to high-resolution retinal imaging can detect early signs of Alzheimer's disease and kidney dysfunction.