vsd-feet-seg

This dataset contains 30 CT scans of the feet. Each scan has a label map of 56 individual bones, with separate labels for left and right. This is the training data of the BoneHub/total-feet-seg model.

Preview of feet segmentations in this dataset.

Source

This dataset combines two releases by M. C. M. Fischer:

What we used	Source
CT images	VSDFullBodyBoneReconstruction (Zenodo)
Bone meshes (MATLAB files)	VSDFullBodyBoneModels v3.0 (Zenodo)

The first release also includes segmentations and meshes, but they are hollow shells, so we did not use them. The filled label maps were created from the closed meshes of the second release.

How the dataset was made

1. Prepare the bone meshes. The bone meshes were converted to one STL file per bone, and the fused foot bones were split into individual bones. These steps and their scripts are described in BoneHub/vsd-lower-extremities-seg.
2. Build the dataset. dataset_creation/create_feet_ct.py uses the lower-leg-and-feet CT scan of each subject and does four things:
   • flips scans that were stored upside down
   • converts the STL meshes into a filled label map on the CT grid
   • crops the CT and the label map around the foot bones, keeping a margin of 100 voxels
   • saves both to feet_ct/

The script contains local paths to the source data, so you need to adjust them before running it. Its requirements are listed in dataset_creation/requirements.txt.

Contents

vsd-feet-seg/
├── feet_ct/
│   ├── dataset.json          # label names and numbers
│   ├── imagesTr/             # 001_0000.nii.gz … 030_0000.nii.gz   CT in Hounsfield units
│   └── labelsTr/             # 001.nii.gz … 030.nii.gz             labels 0–56
├── dataset_creation/         # script used to create the dataset (see above)
└── assets/preview.png

The case numbers match vsd-lower-extremities-seg: case 001 here is the same subject as case 001 there. Subject metadata and bone meshes are available in that repository.

Labels

The label names and numbers are listed in feet_ct/dataset.json and in the model card.

Download

hf download BoneHub/vsd-feet-seg --repo-type dataset --local-dir ./vsd-feet-seg

Training with nnU-Net

feet_ct/ is already in nnU-Net format, but nnU-Net expects dataset folders to be named DatasetXXX_Name. Rename it when you move it into $nnUNet_raw, then train:

mv ./vsd-feet-seg/feet_ct "$nnUNet_raw/Dataset002_vsd-feet-seg"
nnUNetv2_plan_and_preprocess -d 2 -pl nnUNetPlannerResEncL -c 3d_fullres --verify_dataset_integrity
nnUNetv2_train 2 3d_fullres all -tr nnUNetTrainerNoMirroring -p nnUNetResEncUNetLPlans

• The ID (here 002) must not be used by any other dataset or model in your nnU-Net folders. If ID 2 is taken, pick another one and use it with -d and in the train command.
• Use nnUNetTrainerNoMirroring, so that left and right labels are not swapped during training.
• The model card has the full retraining and fine-tuning steps.

Known issues

Two cases are missing some bones, because their source meshes are missing:

• Case 026: the distal phalanges of toes 2, 3 and 4 on the right foot are unlabeled (labels 38, 44, 50).
• Case 030: all phalanges are unlabeled (labels 29–56).

License

This dataset is released under CC BY-NC-SA 4.0, the same license as the source bone models.

Citation

If you use this dataset, please cite both this dataset (doi:10.57967/hf/10470) and the source publication:

@misc{seyedhamidrezaalaviVsdfeetseg,
  title = {Vsd-Feet-Seg},
  author = {Seyed Hamidreza Alavi and Malte Asseln},
  year = {2026},
  publisher = {Hugging Face},
  doi = {10.57967/HF/10470},
  url = {https://huggingface.co/datasets/BoneHub/vsd-feet-seg},
  urldate = {2026-09-16}
}

@article{fischerDatabaseSegmentationsSurface2023,
  title = {Database of Segmentations and Surface Models of Bones of the Entire Lower Body Created from Cadaver {{CT}} Scans},
  author = {Fischer, M.C.M.},
  year = 2023,
  journal = {Scientific Data},
  volume = {10},
  number = {1},
  pages = {763},
  publisher = {Nature Publishing Group},
  issn = {2052-4463},
  doi = {10.1038/s41597-023-02669-z},
}