A deep learning-driven cataract screening model derived from multicenter real-world dataset
Source check: Source checked. Not independent scientific confirmation.
Finding: A cascaded deep-learning system was trained on 22,094 slit-lamp images from 21 ophthalmic institutions across 12 Chinese provinces and municipalities. On an independent test set, a ResNet50-IBN model achieved 93.74% accuracy, 97.74% specificity and AUC 0.953 for cataract classification. The study does not report prospective screening outcomes.
Technology: Cascaded deep-learning screening model; ResNet50-IBN for cataract classification
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Deep-learning models assessed image quality and confounders before classifying cataract and other eye states from slit-lamp images.
- Test and result
- The system assessed image quality, screened confounders and classified cataract, post-surgery, other ocular disease or healthy eyes. ResNet50-IBN reached 93.74% accuracy, 97.74% specificity and AUC 0.953 on the independent test set.
- Limitation / still unknown
- The held-out test is part of the study dataset; effects on prospective screening, referral decisions or patient outcomes are not established.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12695777/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 22 Nov 2026. Review scope: Primary result-bearing material for the stated method, result and evaluation, with bibliographic version information..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.