SegResNet · Whole-body CT

Native MLX weights for MONAI's 104-structure whole-body CT segmentation models. The high-resolution and low-resolution checkpoints correspond to 1.5 mm and 3 mm inference respectively.

The source is MONAI/wholeBody_ct_segmentation at revision 13f6d33ae9de6664571d935ee647039f477c3fad, files models/model.pt and models/model_lowres.pt. The float32 tensors retain their source names and layouts in safetensors. Model architecture and inference are provided by segresnet-mlx.

The models were trained on TotalSegmentator data; see the upstream model card and data attribution. The original MONAI license is included as LICENSE. These weights are converted copies of the source checkpoints.

For research and education. Not validated for clinical use; outputs require human review.

Quick start

uv venv --python 3.13
source .venv/bin/activate

uv pip install "segresnet-mlx>=0.3.3"
uv tool install huggingface-hub

hf download MedMLX/segresnet-wholebody-mlx --revision b52959dc957212571fd4eb4e04ba169e44475152 --local-dir models

Replace the example input paths with your own files.

from segresnet_mlx.medmlx import run

result = run(
    bundle='wholeBody_ct_segmentation',
    weights_path='models/segresnet-wholebody-high.safetensors',
    image_path='ct.nii.gz',
    output_dir='outputs',
)
print(result["outputs"])

For 3 mm inference use bundle='wholeBody_ct_segmentation_lowres' with segresnet-wholebody-low.safetensors.

Citation

If you use MedMLX, please cite the MedMLX paper:

@article{sandoval2026medmlx,
  title   = {MedMLX: On-Device Medical Imaging AI for Apple Silicon},
  author  = {Sandoval, J. and Sandoval, A. and Saleem, A.},
  journal = {medRxiv},
  year    = {2026},
  doi     = {10.64898/2026.10.28.26366762}
}