PedsAutoDeface tool

This tool can be used to automatically deface pediatric brain MRIs (trained on T1w, T2w, T2w-FLAIR, T1w contrast-enhanced). It was trained using the nnU-Net framework on a multi-institutional, heterogeneous dataset (see reference).

Input files can be unprocessed or pre-processed images.

If further training/fine-tuning is desired, the trained model weights can be accessed here: https://drive.google.com/file/d/1P06VrdaMxX_VENOYVRyvFgJN82SMEsMz/view?usp=sharing

If you use this tool in your work, please cite the following reference accordingly:

[...]

STEP 1: Prepare the input files

Organization

Input files must be located in an input/ directory folder (called "input") in NIfTi file format. The exact naming of the files does not matter, the container will process all NIfTi files in the input/ directory separately (the format of the output name of each file will be: [input-file-name]_defaced.nii.gz).

For example input files for a single subject could describe the image type:

input/
    t1ce.nii.gz
    ...

Or include the subject IDs if there are more than one subject:

input/
    sub001_t1.nii.gz
    sub001_t2.nii.gz
    sub001_fl.nii.gz
    sub002_t1.nii.gz
    sub003_t1ce.nii.gz
    ...

STEP 2: Usage

1. Install Docker
2. copy the docker-compose.yml file from this repository into the directory that contains your input/ folder:

   docker-compose.yml
   input/
       sub001_t1.nii.gz
       sub001_t2.nii.gz
       ...
   

3. from within that folder, run the command:

   docker compose up
   

It takes about an hour to fully process 1 MRI file (with 16 GB memory, 2 GHz 4 cores; however, this depends on your machine specs). Defaced images will be stored in an output/ folder with files named [input-file-name]_defaced.nii.gz and the model-predicted face mask as [input-file-name]_face_mask.nii.gz, for example:

input/
    t1ce_defaced.nii.gz
    t1ce_face_mask.nii.gz
    ...

Issues

Please submit any issues you find while using the tool here: [...].