Orbital Wall Segmentation AI (Thin-WallNet)
3D automatic segmentation of the orbital wall
▸ More attributes
Brief overview
This study developed a 3D deep learning segmentation network for automated orbital wall reconstruction from CT imaging, with a focus on difficult thin-wall regions of the orbital floor and medial wall. The model used a modified V-Net architecture with DenseASPP-based multi-scale feature extraction and thin-wall supervision in the loss function. Across 100 retrospectively collected CT scans from a single Chinese center (70 train/30 test), the network achieved Dice scores of 96.1% for the whole orbital wall and 91.5% for thin-wall regions, outperforming conventional thresholding, region-growing, U-Net, and V-Net approaches. Average segmentation time was 4.05 seconds per orbit. No external validation, public code, pretrained weights, calibration analysis, or clinical deployment were reported.
WOW: Dice 96.1%, IoU 92.5%, 95HD 0.51 mm.
WOW: Dice 96.1%, IoU 92.5%, 95HD 0.51 mm. TW: Dice 91.5%, IoU 84.3%, 95HD 0.48 mm. Runtime ≈ 4 seconds per orbit.
| Study design | Methodological/development-only |
|---|---|
| Center type | Single-center |
| Patients | 100 |
| Images / eyes | 200 CT images |
| Split | 70/0/30 |
| 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_xu_2023 |
|---|---|
| Full author list | Zhang, D., Wang, C., Zhou, H., Li, Y., & Chen, X. |
| Journal | Int J Comput Assist Radiol Surg |
| DOI | 10.1007/s11548-023-02924-z |
| Conditions | Orbital wall |
Model internals ▸
| Architecture family | CNN |
|---|---|
| Architecture detail | V-Net–based multi-scale CNN with DenseASPP |
Disclosures ▸
| Funding | Natural Science Foundation of China (81971709; M-0019; 82011530141)' the Foundation of Science and Technology Commission of Shanghai Municipality (20490740700)' Shanghai Jiao Tong University Foundation on Medical and Technological Joint Science Research (YG2021ZD21; YG2021QN72; YG2022QN056; YG2023ZD15; YG2023ZD19); SJTU Global Strategic Partnership Fund (2021 SJTU-HUJI, 2023 SJTU-CORNELL); Cross disciplinary Research Fund of Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine (JYJC202115); Translation Clinical R&D Project of Medical Robot of Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine (IMR-NPH202002); Shanghai Key Clinical Specialty, Shanghai Eye Disease Research Center (2022ZZ01003) |
|---|---|
| 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)