UNesT

UNesT whole-brain segmentation: Singularity container

This record archives UNest.sif, a Singularity container for running UNesT whole-brain segmentation on T1-weighted MRI. The workflow segments 132 anatomical regions and optionally estimates total intracranial volume (TICV) and posterior fossa volume (PFV).

Official source repository:
https://github.com/MASILab/UNesT

MASI Lab documentation:
https://masilab.github.io/brain/unest/

Whole-brain usage documentation:
https://github.com/MASILab/UNesT/blob/main/wholebrainSeg/README.md

RUNNING THE CONTAINER

Use a Linux system with Singularity installed. GPU execution requires an NVIDIA GPU with compatible host drivers.

Place the T1-weighted MRI in an input directory and create writable working and output directories. Replace the example paths below with absolute paths on your system.

Copy the following command as one line:

singularity run -e --contain --home /path/to/inputs -B /path/to/inputs:/INPUTS -B /path/to/working:/WORKING_DIR -B /path/to/outputs:/OUTPUTS -B /tmp:/tmp --nv /path/to/UNest.sif --ticv --w_skull --overlap 0.5 --device 0

DIRECTORY BINDINGS

/INPUTS: input images.

/WORKING_DIR: writable working directory.

/OUTPUTS: output directory.

/tmp: temporary storage.

OPTIONS

--nv enables NVIDIA GPU access.

--device 0 selects GPU 0. The documented default is 0.

--w_skull indicates that the input includes the skull. Omit this option for skull-stripped input.

--ticv enables TICV/PFV estimation alongside anatomical segmentation. Omit this option to run anatomical segmentation alone.

--overlap 0.5 sets sliding-window overlap to 0.5. The documented default is 0.7.

Refer to the whole-brain usage documentation linked above for additional instructions and anatomical label definitions.

CITATIONS AND ATTRIBUTION

The upstream whole-brain documentation requests citation of both publications below. Please also acknowledge the official software repository and cite this Zenodo record to identify the archived container used in your work.

Yu, X., Yang, Q., Zhou, Y., et al. (2023). UNesT: local spatial representation learning with hierarchical transformer for efficient medical segmentation. Medical Image Analysis, article 102939.
https://doi.org/10.1016/j.media.2023.102939

Yu, X., Tang, Y., Yang, Q., Lee, H. H., Bao, S., Huo, Y., and Landman, B. A. (2024). Enhancing hierarchical transformers for whole brain segmentation with intracranial measurements integration. Medical Imaging 2024: Clinical and Biomedical Imaging, Proceedings of SPIE, 12930, 129300K.
https://doi.org/10.1117/12.3009084

Software repository: MASILab. UNesT: Official PyTorch implementation.
https://github.com/MASILab/UNesT

LICENSE AND PROVENANCE

The upstream UNesT source repository is released under the MIT License:
https://github.com/MASILab/UNesT/blob/main/LICENSE

Third-party components bundled in the container remain subject to their respective licenses.