MedMLX/vista3d-ctmr-mlx

NV-Segment-CTMR · MLX weights

Native MLX safetensors weights converted from NVIDIA NV-Segment-CTMR revision 4fb8b4a6b2532be9f1c449a3726fe5440ab4213a, file vista3d_pretrained_model/model.safetensors. automatic.safetensors holds automatic graph tensors; correction.safetensors holds additional point-graph tensors. Names, layouts and float32 values are preserved. These are modified, split copies of NVIDIA's checkpoint.

Use with mlx-segment-ct and checkpoint="NV-Segment-CTMR". The NVIDIA model card states that NV-Segment-CTMR does not support interactive point segmentation. The correction tensors do not establish point-interaction accuracy or support.

Quick start

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

uv pip install "mlx-segment-ct>=0.3.5"
uv tool install huggingface-hub

hf download MedMLX/vista3d-ctmr-mlx --revision e6eb009a242f946572a21610e82160762408bfd5 --local-dir models

Replace the example input paths with your own files.

from mlx_segment_ct.medmlx import run_ctmr

result = run_ctmr(
    image_path='mri.nii.gz',
    weights_path='models',
    output_dir='outputs',
    segment_all=True,
)
print(result["outputs"])

The input can be CT or MRI.

License and use

The model weights are distributed under the NVIDIA OneWay Non-Commercial License. Use is limited to noncommercial research or evaluation, and redistribution is subject to that license. See the original NVIDIA model card for model details and terms.

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

Citation

If you use MedMLX, please cite the MedMLX paper:

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