The abstract is read at the publisher; the summary is JClub's.
In the Rotterdam Study (19,088 eyes; 1.1% perimetric glaucoma), the AIROGS deep-learning algorithm achieved 70.6% sensitivity at 95% specificity with ROC-AUC 0.911.
- Why it matters: External validation in a population-based cohort is needed to confirm that deep-learning glaucoma triage works beyond the training setting and supports scalable screening.
- What they did: The AIROGS model, trained on 101,442 macula-centered CFPs, was externally tested on Rotterdam Study participants aged 50–80, evaluating eye and participant performance with cluster bootstrap CIs and qualitative review of errors.
- The result: Correctly identified true positives had more advanced visual field loss (median mean deviation -10.8 vs -5.8 dB; P < 0.001) and were often flagged 4–7 years before diagnosis, while disc-centered CFPs showed higher discrimination at 90–100% specificity.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.