Skip to content

Latest commit

 

History

History
78 lines (57 loc) · 3.16 KB

README.md

File metadata and controls

78 lines (57 loc) · 3.16 KB

Build Docker image

Use the following command to build the Docker image museg:thigh-model3 (muscle segmentation thigh model version 3):

docker build --rm -f Dockerfile.thigh -t museg:thigh-model3 .

Prerequisites are that the directory contains a copy of the nnU-Net directory nnUNet_trained_models with the trained model, which is not part of this repository due to its size. Further, it is required to install the NVIDIA Container Toolkit.

Run the Docker image

To run the muscle segmentation, the volumes to be segmented need to be put into the correct format. The algorithm requires two volumes per subject named suffix_0000.nii.gz for the in-phase volume at echo time 1.95 ms and suffix_0001.nii.gz for the out-of-phase volume at echo time 2.75 ms, where suffix can be an arbritary string but must be kept equal for both volume files. Both volumes are to be put inside the same directory, e.g., named in. Insided this directory can also be multiple subjects with different suffixes.

.
├── in  # Directory with the input volumes (volumes to be segmented)
│   ├── subject1_0000.nii.gz  # Volume 1 in-phase at echo time 1.95 ms
│   ├── subject1_0001.nii.gz  # Volume 2 out-of-phase at echo time 2.75 ms
│   ├── subject_long_suffix2_0000.nii.gz
│   └── subject_long_suffix2_0001.nii.gz

To infer the thigh muscle segmentation execute within the directory that contains in and out

docker run --rm --gpus all -v "$PWD":/data museg:thigh-model3

The command will run the Docker image museg:thigh-model3 with access to the GPUs (--gpus all) and by mounting the current directory -v "$PWD":/data. It will also remove the contained after it finished (--rm). After successful executing, the segmentations will be saved into a new directory out.

.
├── in
│   ├── subject_0000.nii.gz
│   └── subject_0001.nii.gz
└── out
    ├── subject.nii.gz
    ├── subject_long_suffix2.nii.gz
    ├── plans.pkl
    ├── postprocessing.json
    └── prediction_time.txt

Overwriting directory names

You can adapt the in and out directory names as you wish by specifying the -i and -o parameters. Let's assume you put the data inside mycustomin and want the segmentation to be saved inside mycustomout, execute the Docker by

docker run --rm --gpus all -v "$PWD":/data museg:thigh3 -i ./data/mycustomin -o ./data/mycustomout

Modifying nnU-Net inference parameters

The entry point of the Docker image is the nnU-Net's nnUNet_predict. You can modify any parameters of this function by passing it to the command. For instance, use --disable_tta to speed up inference time by approximately factor 8 by disabling test-time augmentation:

docker run --rm --gpus all -v "$PWD":/data museg:thigh-model3 --disable_tta

Use docker run --rm --gpus all -v "$PWD":/data museg:thigh-model3 --help to learn more about inference parameters.

Inference without a GPU

If no GPU is available, just drop the --gpus all

docker run --rm -v "$PWD":/data museg:thigh-model3