MedSAM2 · CT Lesions

Initialize a CT lesion mask with a box prompt, then propagate it across slices with MedSAM2.

Input	Output
CT + slice-level box prompt	Propagated lesion masks

Apple Silicon · Metal · Python 3.13

Converted from the published MedSAM2 CT lesion checkpoint without retraining.

Quick start

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

uv pip install "medsam2-mlx>=0.2.8"
uv tool install huggingface-hub

hf download MedMLX/medsam2-ct-lesion-mlx --revision 22f5e6dc4850a71abf92ef1a8fea0db14c34d460 --local-dir models

Replace the example input paths with your own files.

from medsam2_mlx.medmlx import run

result = run(
    image_path="ct.nii.gz",
    weights_path="models/MedSAM2_CTLesion.safetensors",
    output_dir="outputs",
    box_xyxy=[411, 270, 442, 301],
    frame_idx=68,
    propagation="both",
    slice_range=[60, 76],
)
print(result["outputs"])

Set the box and slice range for your volume. frame_idx uses the third voxel axis; box coordinates use the first two axes.

Use & license

For research and education. Not validated for clinical use.

The original MedSAM2 model card identifies the checkpoint license as CC-BY-SA-4.0 and separately states that the model weights can only be used for research and education. CC-BY-SA-4.0 permits commercial use; the source does not explain how these statements relate. Both are preserved here for users to review.

Credit the MedSAM2 authors and link the source and license when sharing; identify changes and preserve applicable share-alike terms. MedMLX converted the CT lesion checkpoint to MLX safetensors without retraining. See NOTICE, the license text, and upstream notices.

Paper references

Original model: MedSAM2: Segment Anything in 3D Medical Images and Videos.

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}
}