# MedMLX

> On-device medical imaging AI for Apple Silicon. MedMLX reimplements published medical imaging models in MLX, and all inference runs on your Mac. The inference path uses no PyTorch or CUDA.

MedMLX is research software, not a medical device. Model outputs require human review.

## Install

Requires macOS 14.0 or later on Apple Silicon, Python 3.13 and uv. Install uv:

    curl -LsSf https://astral.sh/uv/install.sh | sh

Install the medmlx command, its MCP server and every model family:

    uv tool install --python 3.13 "medmlx[all,mcp]"

Describe a model's arguments, check readiness, then run it:

    medmlx describe mlx-segment-ct
    medmlx readiness mlx-segment-ct --arguments arguments.json --root /path/to/work
    medmlx run mlx-segment-ct --arguments arguments.json --root /path/to/work

Download weights ahead of time for offline use:

    medmlx setup --all --progress

## Agents (MCP)

An MCP client can list, check and run models. Configure this as its stdio command, with a folder that already exists:

    medmlx serve --root /path/to/work

- [MCP guide](https://medmlx.org/mcp/): install, client setup, job workflow and tool reference
- [Command line](https://medmlx.org/cli/): the medmlx command, its subcommands and weight downloads
- [Python API](https://medmlx.org/python/): run, describe_model, model_readiness and the other public functions
- [Model catalog as JSON](https://medmlx.org/catalog.json): every runtime with its version, links and install command

## Maintainers and citation

MedMLX is maintained by [Joseph Sandoval](https://josephsandoval.org/) and [Alfredo Sandoval Jr.](https://alfredo-sandoval.com/) in the [AI-IR lab](https://ai-ir.org/) at the University of Texas Medical Branch.

When you use or mention MedMLX, cite the article:

Sandoval J, Sandoval A Jr, Saleem A. MedMLX: On-Device Medical Imaging AI for Apple Silicon. medRxiv. 2026. https://doi.org/10.64898/2026.10.28.26366762

    @article{sandoval2026medmlx,
      title   = {MedMLX: On-Device Medical Imaging AI for Apple Silicon},
      author  = {Sandoval, Joseph and Sandoval, Jr., Alfredo and Saleem, Arsalan},
      journal = {medRxiv},
      year    = {2026},
      doi     = {10.64898/2026.10.28.26366762},
      url     = {https://doi.org/10.64898/2026.10.28.26366762}
    }

## Docs

- [Get started](https://medmlx.org/get-started/): install MedMLX and run a model
- [Models](https://medmlx.org/models/): all 19 runtimes by task
- [Community](https://medmlx.org/community/): model requests, questions, bug reports and shared work
- [Article](https://doi.org/10.64898/2026.10.28.26366762): Joseph Sandoval, Alfredo Sandoval Jr., Arsalan Saleem. MedMLX: On-Device Medical Imaging AI for Apple Silicon. medRxiv, 2026.

## Models

### Segmentation

- [nnU-Net](https://medmlx.org/models/nnunet-mlx/): Segmentation of liver and pancreas CT, brain tumor MRI and PET/CT lesions.
  Package nnunet-mlx 0.6.4. [PyPI](https://pypi.org/project/nnunet-mlx/) · [Source](https://github.com/MedMLX/nnunet-mlx/tree/v0.6.4)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "nnunet-mlx>=0.6.4"

- [TotalSegmentator](https://medmlx.org/models/totalsegmentator-mlx/): Identify 117 organs and other body structures in CT scans.
  Package totalsegmentator-mlx 0.1.6. [PyPI](https://pypi.org/project/totalsegmentator-mlx/) · [Source](https://github.com/MedMLX/totalsegmentator-mlx/tree/v0.1.6)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "totalsegmentator-mlx>=0.1.6"

- [MedSAM2](https://medmlx.org/models/medsam2-mlx/): Lesion or organ segmentation from a box, point or mask prompt, tracked through the volume.
  Package medsam2-mlx 0.2.8. [PyPI](https://pypi.org/project/medsam2-mlx/) · [Source](https://github.com/MedMLX/medsam2-mlx/tree/v0.2.8)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "medsam2-mlx>=0.2.8"

- [BioMedParse](https://medmlx.org/models/biomedparse-mlx/): Text-prompted 3D segmentation with Microsoft BioMedParse v2.
  Package biomedparse-mlx 0.1.5. [PyPI](https://pypi.org/project/biomedparse-mlx/) · [Source](https://github.com/MedMLX/biomedparse-mlx/tree/v0.1.5)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "biomedparse-mlx>=0.1.5"

- [NV-Segment-CT](https://medmlx.org/models/mlx-segment-ct/): Automatic and prompted 3D CT segmentation with NVIDIA VISTA-3D.
  Package mlx-segment-ct 0.3.5. [PyPI](https://pypi.org/project/mlx-segment-ct/) · [Source](https://github.com/MedMLX/mlx-segment-ct/tree/v0.3.5)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "mlx-segment-ct>=0.3.5"

- [SegResNet](https://medmlx.org/models/segresnet-mlx/): Segmentation of brain tumors, kidney structures and whole-body anatomy from MRI or CT.
  Package segresnet-mlx 0.3.3. [PyPI](https://pypi.org/project/segresnet-mlx/) · [Source](https://github.com/MedMLX/segresnet-mlx/tree/v0.3.3)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "segresnet-mlx>=0.3.3"

- [UNesT](https://medmlx.org/models/unest-mlx/): Kidney cortex, medulla and collecting system segmentation in contrast-enhanced CT.
  Package unest-mlx 0.2.7. [PyPI](https://pypi.org/project/unest-mlx/) · [Source](https://github.com/MedMLX/unest-mlx/tree/v0.2.7)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "unest-mlx>=0.2.7"

- [SwinUNETR](https://medmlx.org/models/swin-unetr-mlx/): Organ segmentation in abdominal CT.
  Package swin-unetr-mlx 0.2.6. [PyPI](https://pypi.org/project/swin-unetr-mlx/) · [Source](https://github.com/MedMLX/swin-unetr-mlx/tree/v0.2.6)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "swin-unetr-mlx>=0.2.6"

- [SynthSeg](https://medmlx.org/models/synthseg-mlx/): Brain segmentation and parcellation.
  Package synthseg-mlx 0.3.4. [PyPI](https://pypi.org/project/synthseg-mlx/) · [Source](https://github.com/MedMLX/synthseg-mlx/tree/v0.3.4)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "synthseg-mlx>=0.3.4"

- [FleXray](https://medmlx.org/models/flexray-mlx/): Radiograph segmentation.
  Package flexray-mlx 0.2.6. [PyPI](https://pypi.org/project/flexray-mlx/) · [Source](https://github.com/MedMLX/flexray-mlx/tree/v0.2.6)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "flexray-mlx>=0.2.6"

- [FlexibleUNet](https://medmlx.org/models/flexible-unet-mlx/): Identify surgical instruments in endoscopy images.
  Package flexible-unet-mlx 0.2.6. [PyPI](https://pypi.org/project/flexible-unet-mlx/) · [Source](https://github.com/MedMLX/flexible-unet-mlx/tree/v0.2.6)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "flexible-unet-mlx>=0.2.6"

### Detection

- [3D RetinaNet](https://medmlx.org/models/lung-nodule-mlx/): MONAI RetinaNet lung nodule detection in 3D chest CT.
  Package lung-nodule-mlx 0.2.6. [PyPI](https://pypi.org/project/lung-nodule-mlx/) · [Source](https://github.com/MedMLX/lung-nodule-mlx/tree/v0.2.6)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "lung-nodule-mlx>=0.2.6"

### Generation

- [FDG PET/CT](https://medmlx.org/models/fdg-pet-ct-mlx/): Synthetic FDG PET from CT and organ labels, with scanner blur, scatter and count noise.
  Package fdg-pet-ct-mlx 0.2.6. [PyPI](https://pypi.org/project/fdg-pet-ct-mlx/) · [Source](https://github.com/MedMLX/fdg-pet-ct-mlx/tree/v0.2.6)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "fdg-pet-ct-mlx>=0.2.6"

- [NV-Generate](https://medmlx.org/models/mlx-generate-ctmr/): Synthetic 3D CT and MRI generation with NVIDIA NV-Generate (MAISI).
  Package mlx-generate-ctmr 0.5.2.post2. [PyPI](https://pypi.org/project/mlx-generate-ctmr/) · [Source](https://github.com/MedMLX/mlx-generate-ctmr/tree/v0.5.2.post2)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "mlx-generate-ctmr>=0.5.2.post2"

- [LesionDiffusion](https://medmlx.org/models/lesiondiffusion-mlx/): Synthetic lesions in CT, with control of organ, size, placement and phenotype.
  Package lesiondiffusion-mlx 0.2.6.post2. [PyPI](https://pypi.org/project/lesiondiffusion-mlx/) · [Source](https://github.com/MedMLX/lesiondiffusion-mlx/tree/v0.2.6.post2)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "lesiondiffusion-mlx>=0.2.6.post2"

### Vision-language

- [MLX-Reason-CT](https://medmlx.org/models/mlx-reason-ct/): Writes a structured report for a chest or abdomen CT volume, or answers a question about it.
  Package mlx-reason-ct 0.3.6. [PyPI](https://pypi.org/project/mlx-reason-ct/) · [Source](https://github.com/MedMLX/mlx-reason-ct/tree/v0.3.6)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "mlx-reason-ct>=0.3.6"

- [MedGemma](https://medmlx.org/models/med-vlm-mlx/): Ask questions about medical images and receive written responses on your Mac.
  Package med-vlm-mlx 0.3.2. [PyPI](https://pypi.org/project/med-vlm-mlx/) · [Source](https://github.com/MedMLX/med-vlm-mlx/tree/v0.3.2)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "med-vlm-mlx>=0.3.2"

### Quantitative imaging

- [PyRadiomics](https://medmlx.org/models/mlxradiomics/): Measure image intensity, texture and shape within a selected region.
  Package mlxradiomics 0.1.0.dev10. [PyPI](https://pypi.org/project/mlxradiomics/) · [Source](https://github.com/MedMLX/mlx-radiomics/tree/v0.1.0.dev10)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "mlxradiomics>=0.1.0.dev10"

### Feature extraction

- [RadImageNet](https://medmlx.org/models/radimagenet-mlx/): Extract numerical image features for research with models trained on RadImageNet.
  Package radimagenet-mlx 0.3.3. [PyPI](https://pypi.org/project/radimagenet-mlx/) · [Source](https://github.com/MedMLX/radimagenet-mlx/tree/v0.3.3)
  Install on its own in a new environment:

        uv venv --python 3.13
        source .venv/bin/activate
        uv pip install "radimagenet-mlx>=0.3.3"

## Optional

- [About](https://medmlx.org/about/): project, developer links, intended use and licenses
- [Image credits](https://medmlx.org/assets/hero/credits.html): datasets and licenses for the homepage media
- [Source on GitHub](https://github.com/MedMLX/medmlx)
- [medmlx on PyPI](https://pypi.org/project/medmlx/)
- [Weights on Hugging Face](https://huggingface.co/MedMLX)