Minimal annotation muscle segmentation (MuSeg) based on artificial intelligence (AI), short MuSegAI. If you use this work, please consider citing it.

The repository provides a command line interface (CLI) based on a Docker image for segmentation. Prerequisites for the segmentation are two Dixon MRI volumes at echo times 1.95 and 2.75 ms (in- and out-of-phase volumes). Currently, a segmentation model for the thigh muscles is available, which segments 13 muscles (see ./docker/labels_thigh.txt). The volumes can either contain both thighs or left and right only.

Install the MuSeg-AI package by (tested on Ubuntu 22.04 with Python 3.10)

git clone https://github.com/fabianbalsiger/minimal-annotation-muscle-segmentation.git
cd minimal-annotation-muscle-segmentation
pip install .[cli]

You can also directly work with the Docker image available on DockerHub, pull it by

docker pull fabianbalsiger/museg:thigh-model3

Follow the docker/README.md instructions on how to execute a Docker container for segmentation.

Let's assume having volumes saved in a directory in as volume195.mha and volume275.mha (the names can be arbritary). Segment them by

museg-ai in/volume195.mha in/volume275.mha

The files volume195.mha, the segmentation, and labels.txt, an ITK-SNAP labels file, will be saved at the location where the script has been executed.

NOTE: If the Docker image for segmentation has not yet been pulled, it will be done automatically, which might take a while.

Print all available options by

museg-ai --help

If you use this work, please cite:

P.-Y. Baudin*, F. Balsiger*, L. Beck, J.-M. Boisserie, S. Jouan, B. Marty, H. Reyngoudt, O. Scheidegger, "A Minimal Annotation Pipeline for Deep Learning Segmentation of Skeletal Muscles", NMR in Biomedicine 38, no. 7 (2025): e70066, https://doi.org/10.1002/nbm.70066.

@article{BaudinBalsiger2025a,
author = {Baudin, Pierre-Yves and Balsiger, Fabian and Beck, Lea and Boisserie, Jean-Marc and Jouan, Sophie and Marty, Benjamin and Reyngoudt, Harmen and Scheidegger, Olivier},
title = {A Minimal Annotation Pipeline for Deep Learning Segmentation of Skeletal Muscles},
journal = {NMR in Biomedicine},
volume = {38},
number = {7},
pages = {e70066},
keywords = {automatic segmentation, neuromuscular disorders, nnU-net, quantitative MRI, skeletal muscles},
doi = {https://doi.org/10.1002/nbm.70066},
url = {https://analyticalsciencejournals.onlinelibrary.wiley.com/doi/abs/10.1002/nbm.70066},
year = {2025}

• Pierre-Yves Baudin and Fabian Balsiger contributed equally to this work.

A little project cheatsheet for development:

• pre-commit: pre-commit run --all-files --hook-stage=manual
• pytest: pytest or pytest -s
• coverage: coverage run -m pytest followed by coverage html or coverage report
• create towncrier entry: towncrier create 123.added --edit

Follow these steps for the initial project setup as well as setting up a new development environment:

1. See docs/getting_started.md or docs/quickstart.md for how to get up & running.
2. Check docs/project_specific_setup.md for project specific setup.
3. See docs/detect_secrets.md for more on creating a .secrets.baseline file using detect-secrets.
4. See docs/using_towncrier.md for how to update the CHANGELOG.md using towncrier.