Blepharoptosis AI Platform (DenseNet121)
Binary classification of blepharoptosis from single-eye crops of clinical photographs
▸ More attributes
Brief overview
This study developed a convolutional neural network (CNN) system to identify blepharoptosis from standard clinical photographs without manual annotation or external reference markers. Using 434 retrospectively collected single-eye images from a tertiary ophthalmic center in Taiwan, the authors compared 22 CNN configurations including DenseNet, ResNet, VGG, AlexNet, and SqueezeNet architectures with and without ImageNet pretraining. The best-performing model, DenseNet121 without pretraining, achieved 90.1% sensitivity, 82.4% specificity, and AUROC 0.95 on internally validated test data using 5-fold cross-validation. Images were labeled by consensus of oculoplastic surgeons. No external validation, prospective deployment, open-source code, or public model weights were reported. The dataset consisted entirely of Asian patients, limiting generalizability.
Best model (DenseNet121, no pretraining): sensitivity 0.90; specificity 0.82; accuracy 0.89; AUROC 0.95.
Best model (DenseNet121, no pretraining): sensitivity 0.90; specificity 0.82; accuracy 0.89; AUROC 0.95. Pretrained models trained faster but offered similar accuracy.
| Study design | Diagnostic accuracy study |
|---|---|
| Center type | Single-center |
| Patients | Not reported |
| Images / eyes | 434 images |
| Split | ~72/18/10 + 5-fold cross-validation |
| Reporting frameworks | None stated |
| Age range | >18 |
|---|---|
| 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_2021 |
|---|---|
| Full author list | Perera, C., Chen, K. W., Myung, D., Chiu, H. K., Fuh, C. S., Hsu, C. R., Liao, S. L., & Kossler, A. L. |
| Journal | Int J Med Inform |
| DOI | 10.1016/j.ijmedinf.2021.104402 |
| Conditions | Blepharoptosis |
Model internals ▸
| Architecture family | CNN |
|---|---|
| Architecture detail | Multiple CNN architectures (best: DenseNet121). |
Disclosures ▸
| Funding | National Eye Institute (P30 EY026877); Research to Prevent Blindness (RPB); Computational and storage resources were sponsored by Taiwan National Center for High-performance Computing (NCHC) |
|---|---|
| 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)