lung-nodule-mlx

Detect lung nodule candidates in 3D chest CT with MONAI RetinaNet and native MLX.

Docs · MedMLX

A 3D nodule candidate localized in axial, coronal and sagittal CT views.

Apple Silicon · Metal · Python 3.13 / 3.14 · uv

Quick start

uv venv --python 3.13
source .venv/bin/activate
repo="https://github.com/MedMLX/lung-nodule-mlx"
uv pip install \
  "lung-nodule-mlx[conversion] @ git+$repo" \
  huggingface-hub

Download and convert the checkpoint once, then run:

python -c 'from pathlib import Path; from huggingface_hub import hf_hub_download; import lung_nodule_mlx as m; m.convert_weights(Path(hf_hub_download("MONAI/lung_nodule_ct_detection", "models/model.pt")), Path("network.safetensors"))'

c++ -std=c++17 -O2 -dynamiclib \
  "$(python -c 'import lung_nodule_mlx, pathlib; print(pathlib.Path(lung_nodule_mlx.__file__).with_name("selection.cpp"))')" \
  -o selection.dylib

from lung_nodule_mlx.medmlx import run

result = run(
    image_path="ct.nii.gz",
    weights_path="network.safetensors",
    selector_path="selection.dylib",
    output_dir="out",
    resample_raw_input=True,
)
print(result["outputs"]["detections"])

Requires Apple’s command-line developer tools for box selection. Keep resample_raw_input=True for ordinary CT.

License

Apache-2.0. See third-party notices. Model weights retain their publisher terms.

For research and education. Outputs require human review.