Referable Blepharoptosis AI (VGG16 CNN)
Automated detection of referable blepharoptosis (true ptosis + pseudoptosis) from single-eye external photographs
▸ More attributes
Brief overview
A VGG-16–based convolutional neural network was developed to identify referable blepharoptosis, including true ptosis and pseudoptosis, from cropped periocular photographs. Using 782 retrospectively collected single-eye images from a tertiary oculoplastic clinic, the model achieved 90% accuracy, 92% sensitivity, 88% specificity, and AUC 0.987 on an internal test set, outperforming non-ophthalmic physicians. Grad-CAM visualization demonstrated model attention around the upper eyelid margin and corneal light reflex, consistent with MRD1 assessment principles. Transfer learning from ImageNet and standard augmentation techniques were applied during training. No external validation, public code, pretrained weights, or clinical deployment were reported. The dataset consisted exclusively of adult Asian patients, limiting generalizability.
On a 50-image test set: accuracy 0.90; sensitivity 0.92; specificity 0.88; AUC 0.99. Model outperformed three non-ophthalmic physicians (mea
On a 50-image test set: accuracy 0.90; sensitivity 0.92; specificity 0.88; AUC 0.99. Model outperformed three non-ophthalmic physicians (mean accuracy 0.77).
| Study design | Diagnostic accuracy study |
|---|---|
| Center type | Single-center |
| Patients | Not reported |
| Images / eyes | 782 images |
| Split | ~75/19/6 |
| Reporting frameworks | None stated |
| Age range | >20 |
|---|---|
| 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_hung_2022 |
|---|---|
| Full author list | Chen, K. W., Perera, C., Chiu, H. K., Hsu, C. R., Myung, D., Luo, A. C., Fuh, C. S., Liao, S. L., & Kossler, A. L. |
| Journal | J Pers Med |
| DOI | 10.3390/jpm12020283 |
| Conditions | Blepharoptosis |
Model internals ▸
| Architecture family | CNN |
|---|---|
| Architecture detail | VGG-16–based CNN |
Disclosures ▸
| Funding | Research to Prevent Blindness and the National Eye Institute (P30-026877) |
|---|---|
| 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)