Strabismus Surgical Planning AI Platform
Strabismus screening (classification), deviation angle estimation, and exotropia surgical target angle prediction
▸ More attributes
Brief overview
Mao and colleagues developed a three-headed Inception-ResNet-V2 platform that screens for strabismus, estimates deviation angle, and predicts the exotropia surgical target angle, all from corneal light-reflection photographs. The model was trained on 2,154 patients retrospectively and validated prospectively on 323 patients at the same center, achieving AUC 0.98 for screening, deviation-angle correlation r 0.95–0.98 against the prism cover test, and surgical-target-angle correlation r 0.76–0.86. It is the highest-quality oculoplastic AI study in the eyemodelhub corpus to date by APPRAISE-AI score, principally because of the prospective validation and the breadth of clinically relevant tasks evaluated.
Screening: AUC 1.00 (retrospective) / 0.98 (prospective); accuracy 99% / 97%.
Screening: AUC 1.00 (retrospective) / 0.98 (prospective); accuracy 99% / 97%. Deviation regression: r=0.95–0.98; mean error ~2.6–2.9°. Surgical target angle: r=0.76–0.86; mean error ~2.3–2.5°.
| Study design | Diagnostic accuracy study |
|---|---|
| Center type | Single-center |
| Patients | 3279 |
| Images / eyes | 5,832 images |
| Split | 70/15/15 + retrospective hold-out test set + prospective same-center hold-out test set |
| Reporting frameworks | None stated |
| Age range | Retrospective: mean 13.4 Prospective: mean 14.8 |
|---|---|
| Sex (% female) | Not reported |
| Race / ethnicity reported | N |
| Subgroup performance | N |
- Concept
- Prototype
- Internal val.
- External val.
- Regulator
- Clinical
▸ Why this score?
The APPRAISE-AI tool grades AI/ML medical studies across 24 items in 6 subdomains, scored independently by two reviewers. The total reflects the methodological transparency of the publication — not the underlying clinical value of the model.
How to read this: high scores in clinical relevance and reporting with low scores in methodological conduct, robustness, or reproducibility is the dominant pattern in ophthalmic AI today.
Quality bands: Low <40 · Moderate 40–59 · High 60–79 · Very High 80+. Most published ophthalmic AI sits in the Low–Moderate band; studies that break into High typically combine multi-center datasets, prospective evaluation, and open-source release.
Bibliographic record ▸
| Model ID | oculoplastics_mao_2021 |
|---|---|
| Full author list | Yang, Y., Guo, C., Zhu, Y., Chen, C., Chen, J., Liu, L., Chen, L., Mo, Z., Lin, B., Zhang, X., Li, S., Lin, X., & Lin, H. |
| Journal | Ann Transl Med |
| DOI | 10.21037/atm-20-5442 |
| Conditions | Strabismus |
Model internals ▸
| Architecture family | CNN |
|---|---|
| Architecture detail | Inception-ResNet-V2 CNN with three output heads |
Disclosures ▸
| Funding | National Key Research and Development Program of China (2018YFC0116500); the National Natural Science Foundation of China (81770967); the Science and Technology Planning Projects of Guangdong Province (2017B030314025, 2018B010109008, 2019B030316012); Guangdong Science and Technology Innovation Leading Talents (2017TX04R031) |
|---|---|
| Conflicts of interest | No conflicts |
Status, version, reviewers ▸
| Card status | Published |
|---|---|
| Card version | v1.0.0 |
| First added | 2026-04-28 |
| Last reviewed | 2026-04-28 |
| Reviewers | attribution opt-in pending |
Card changelog ▸
- 2026-04-28 initial publication (v1.0.0)