DentalSegmentator-KonfAI
KonfAI adaptation of DentalSegmentator β dento-maxillo-facial CT / CBCT segmentation, built with KonfAI.
π§© Model
| Task | Modality | Labels | Ensemble |
|---|---|---|---|
DentalSegmentator |
CT / CBCT | 5 | 1 |
Labels: 1 Maxilla & Upper Skull Β· 2 Mandible Β· 3 Upper Teeth Β· 4 Lower Teeth Β· 5 Mandibular canal.
nnU-Net v2 3d_fullres PlainConvUNet (Dataset111_453CT, fold 0), converted weight-exact by build.py Β·
patch [160, 128, 112] Β· resampled to 0.45 Γ 0.45 Γ 0.31 mm.
π Usage
- Generic runner:
konfai-apps infer VBoussot/DentalSegmentator-KonfAI:DentalSegmentator -i input_cbct.nii.gz -o output/ - Interactive: SlicerKonfAI
- TTA / uncertainty: up to 7 random mirror copies (
--tta 7; the model was trained with mirroring on all axes),-uncertaintyfor the per-voxel label variance.
β‘ Performance
Slicer's PostDentalSurgery CBCT (360 Γ 360 Γ 330, 0.5 mm), same weights, same PyTorch build (2.12.1, cu13.0),
single NVIDIA RTX PRO 5000 (24 GB), nnU-Net 2.x without TTA (2026-09-30).
| Tool | Time | Peak RAM |
|---|---|---|
| KonfAI | 15.9 s | 2.3 GB |
| Original (nnU-Net) | 60.7 s | 6.4 GB |
Dice against the original's output: 0.976 / 0.985 / 0.984 / 0.982 / 0.950 (labels 1β5), the same spread as the original with vs. without its mirroring TTA (0.988 / 0.985 / 0.988 / 0.982 / 0.934).
π Citation
Dot G, et al. DentalSegmentator: robust open source deep learning-based CT and CBCT image segmentation. Journal of Dentistry (2024) doi:10.1016/j.jdent.2024.105130
Isensee F, et al. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat Methods. 2021;18(2):203-211. doi:10.1038/s41592-020-01008-z
βοΈ License
The weights are DentalSegmentator's nnU-Net parameters by Dot et al., taken from the
v1.0.0-alpha release
and converted to KonfAI's format without retraining. The authors publish this model on
Zenodo under CC BY 4.0, so the converted weights are shared here
under the same CC BY 4.0 license: cite the papers above when
you use them. The configuration files and build.py are Apache-2.0.
π Links
- π§ Original DentalSegmentator: github.com/gaudot/SlicerDentalSegmentator (Apache-2.0) β the app icon is its official logo.
- π§ KonfAI: github.com/vboussot/KonfAI