AryanNsc/dental-nvsegment-ct-core

Dental NV-Segment-CT Core

Fine-tuned NV-Segment-CT checkpoint for dental CBCT core anatomy segmentation.

This model adapts NVIDIA's CT segmentation workflow toward dental CBCT anatomy, with a focus on mandible, maxilla, mandibular canal, and teeth region segmentation.

Research use only. Not for clinical diagnosis, treatment planning, or autonomous medical decision-making.

Demo

Companion GitHub repository:

Guney-olu/dental-cbct-segmentation-research

Labels

ID	Label
0	background
1	mandible
2	maxilla
3	mandibular canal
4	teeth-all
5	implant, experimental
6	lesion, experimental

Core anatomy mean Dice is computed over:

• mandible
• maxilla
• mandibular canal
• teeth-all

Implant and lesion classes are included as experimental labels and should not be treated as reliable clinical outputs.

Model Artifact

Expected checkpoint file:

model.pt

Intended Use

This checkpoint is intended for research experiments involving:

• dental CBCT anatomy segmentation
• 2D/3D visualization of predicted anatomy
• segmentation baseline comparison
• mask and mesh generation for research workflows
• future synthetic-data and mask-conditioned generation experiments

How To Use

Inference code, demo viewer, and mesh/overlay export scripts are available in the companion GitHub repository:

git clone https://github.com/Guney-olu/dental-cbct-segmentation-research.git

The repository includes scripts for:

• downloading/loading this checkpoint
• running local inference through NV-Segment-CT
• generating 2D overlays
• exporting 3D GLB meshes
• serving a browser-based demo

Base Model

This work is derived from:

nvidia/NV-Segment-CT

Please review NVIDIA's model license and usage terms before use, redistribution, or derivative deployment.

Limitations

• Research use only.
• Not validated for clinical use.
• Not a medical device.
• Performance may vary across CBCT scanners, acquisition protocols, field of view, artifacts, and patient anatomy.
• Implant and lesion labels are experimental.
• Human expert review is required for any medically relevant interpretation.