Performance of Artificial Intelligence-Based Models for Epiretinal Membrane Diagnosis: A Systematic Review and Meta-Analysis
Bibliographic record
Abstract
TOPIC: Epiretinal membrane (ERM) can impair central vision by forming a pre-retinal fibrous layer on the inner retina. Artificial intelligence (AI)-based tools may streamline ERM diagnosis, but their overall performance and factors affecting accuracy require evaluation. CLINICAL RELEVANCE: With an aging population, ERM prevalence is expected to rise, placing increased demands on clinical resources. Early detection via AI models could expedite diagnosis, reduce subjective errors, and guide timely surgical intervention. This systematic review and meta-analysis evaluates the pooled diagnostic performance of AI models for detecting ERM and identifies study- and model-level factors influencing their performance. DESIGN: Systematic review and meta-analysis. METHODS: Comprehensive searches were conducted in Medline, Embase, Cochrane Library, Web of Science, and preprint databases from inception to June 2024. Included studies evaluated AI models for ERM diagnosis. Study quality and risk of bias were assessed using the Quality Assessment for Diagnostic Accuracy Studies 2 (QUADAS-2) tool. A random-effects model was applied to pool diagnostic accuracy, sensitivity, specificity, and diagnostic odds ratio. Subgroup analyses explored factors affecting model performance. The study protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO - CRD42024563571). RESULTS: Of 379 articles screened, 26 met inclusion criteria, and 19 contributed to the meta-analysis. Study settings were predominantly hospital-based (76.9%), with some studies from academic computer and biomedical science departments (15.4%) and community centers (7.7%). Quality assessments suggested low or unclear risk of bias and applicability concerns in 95% of studies. The pooled sensitivity was 90.1% (95% CI: 85.8-93.2), and the pooled specificity was 95.7% (95% CI: 88.8-95.2). Subgroup analysis showed higher specificity (97.1%, 95% CI: 96.0-97.9) in AI models using color fundus photographs than optical coherence tomography scans, which had a specificity of 92.6% (95% CI: 88.8-95.2). External validation was performed in 26.9% of studies. All included studies used expert human grading as the reference standard, of which 25 (96.2%) were based on the same imaging modality as the AI input. The proportion of ERM cases in development datasets varied across studies, particularly between single-disease and multiclass models. CONCLUSIONS: AI models demonstrate high diagnostic performance for ERM. However, limited external validation and variability in AI development methodologies limits direct comparison between models and real-world applicability. Future work should standardize model development and reporting practices, improve data interoperability, and develop prediction models to track disease progression and determine optimal surgical timing.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.014 | 0.035 |
| Meta-epidemiology (narrow) | 0.003 | 0.002 |
| Meta-epidemiology (broad) | 0.016 | 0.039 |
| Bibliometrics | 0.003 | 0.003 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.003 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".