PeriorbitAI
Automated quantitative measurement of eyelid and periorbital landmarks (MRD1, MRD2, brow and canthal heights)
▸ More attributes
Brief overview
PeriorbitAI uses a ResNet-50 encoder and U-Net decoder to segment eyelid, brow, and canthal landmarks from standard external facial photographs and derive linear measurements (MRD1, MRD2, brow heights). Across 418 prospectively validated images at a single US center, automated measurements achieved Dice 0.82–0.96 with intraclass correlation 0.90–0.95 against human graders and mean absolute error of roughly 0.5–2 mm. Pretrained weights and code are openly available on GitHub. The model has not been validated externally on multi-institutional cohorts and was not deployed clinically. It is the most code-transparent oculoplastic measurement model in the literature to date.
Dice coefficients 0.82–0.96; mean absolute error ~0.5–2 mm vs human graders; ICC 0.90–0.95.
Dice coefficients 0.82–0.96; mean absolute error ~0.5–2 mm vs human graders; ICC 0.90–0.95.
| Study design | Prospective cohort |
|---|---|
| Center type | Single-center |
| Patients | Not reported |
| Images / eyes | 418 images |
| Split | 80/20 + retrospective hold-out test set + prospective hold-out test set |
| Reporting frameworks | None stated |
| Age range | Not reported |
|---|---|
| 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_vanbrummen_2021 |
|---|---|
| Full author list | Owen, J. P., Spaide, T., Froines, C., Lu, R., Lacy, M., Blazes, M., Li, E., Lee, C. S., Lee, A. Y., & Zhang, M. |
| Journal | Am J Ophthalmol |
| DOI | 10.1016/j.ajo.2021.05.007 |
| Conditions | Periorbital measurement |
Model internals ▸
| Architecture family | CNN |
|---|---|
| Architecture detail | Convolutional neural network (ResNet-50 encoder and U-Net-style decoder) |
Disclosures ▸
| Funding | NIH/NIA R01AG060942; NIH/NEI K23EY029246, Research to Prevent Blindness Career Development Award; Latham Vision Innovation Award; unrestricted grant from Research to Prevent Blindness |
|---|---|
| Conflicts of interest | Aaron Y. Lee reports support from the US Food and Drug Administration, grants from Santen, Regeneron, Carl Zeiss Meditec, and Novartis, personal fees from Genentech, Topcon, and Verana Health, outside of the submitted work |
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)