Pediatric automated multi-parametric skull stripping with nnUNet

This pipeline can be used to generate AI-predicted brain masks and skull-stripped images for pediatric patients with multi-parametric MRIs. It was trained using the nnU-Net framework on a multi-institutional, heterogeneous dataset.

Dependencies include:

1. Python 3.9
2. PyTorch
3. nnUNet v1

The package will run nnUNet testing/inference with the pre-trained auto-skull-stripping model on the input files.

Acknowledgement

Ariana Familiar, PhD, Center for Data-Driven Discovery in Biomedicine (D3b), Children's Hospital of Philadelphia

STEP 1: Prepare the input files

Required inputs: Following pre-processed multi-parametric pediatric brain MRI scans:

1. T1-weighted pre-contrast (T1w)
2. T1-weighted post-contrast (T1w post-contrast)
3. T2-weighted (T2w)
4. T2-weighted FLAIR (T2w-FLAIR)

Input files (raw data) must be located in an directory folder and named with the following format: [subID]_[imageID]...[.nii/.nii.gz] where the imageID for each image type is:

Image type	imageID	nnUNet naming
T2w-FLAIR	FL	0000
T1w	T1	0001
T1w post-contrast	T1CE	0002
T2w	T2	0003

NOTE: the exact file format is required with an underscore: [subID]_[imageID]

For example:

input/
    sub001_FL.nii.gz
    sub001_T1.nii.gz
    sub001_T1CE.nii.gz
    sub001_T2.nii.gz
    sub002_FL.nii.gz
    ...

Configured to run on CPU.

STEP 2: Usage

Install Docker

Copy all files into a single directory

Push to Docker Hub

Build the image locally:

docker build -t afam00/peds-brain-auto-skull-strip:0.0.0 .

Push the image to the Docker Hub:

docker image push afam00/peds-brain-auto-skull-strip:0.0.0

Running Inference

From within the directory:

docker build -t peds-brain-auto-skull-strip .

docker run --rm peds-brain-auto-skull-strip

Available models:

• nnUNet-based skull-stripping using multi-parametric brain MRI scans as input: Version 1

References

• Isensee, F., Jaeger, P.F., Kohl, S.A.A. et al. "nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation." Nat Methods (2020). https://doi.org/10.1038/s41592-020-01008-z