Tested on x86/64 Linux-based system.

The project is used to do preprocessing on brain MR images (.nii files).

There is a pipeline include those features:

• Dockerize the environment
• Auto AC-PC detection & alignment
• Registration
• Skull Stripping
• Segmentation (using K-means to split GM, WM and CSF)
• Visualize the results (using nilearn)

All the features implemented as nipype's interface are connected in a workflow.

In this repo, the dataset is downloaded from LONI Image Data Archive (IDA). Collect and download AD and NC screening sample of ADNI1 and ADNI2, and extract them into this folder.

You can just place the .nii samples in ./data.

If you only have DICOM files, you can use DICOM to NIfTI Online Converter to convert them into NIfTI format.

For example, folder ./data structure is like this:

./data
├── 099_S_4206.nii
└── 099_S_4205.nii

0 directories, 2 files

1. Build Neuro Docker image

docker build --tag neuro:latest --file Dockerfile.neuro .

2. Build custom Docker image

We need to install custom tools in the Docker image.

docker build --tag neuro_custom:latest --file Dockerfile .

3. Run the workflow

docker run --rm -it \
    --workdir /src \
    --volume ./src:/src \
    --volume ./utils:/utils \
    --volume ./data:/data \
    --volume ./output:/output \
    --name neuro_workflow \
    neuro_custom python workflow.py

Segmentation results are shown as cover.

When the workflow ran successfully, all the results of each step will be saved in ./output.

And the workflow graph will be saved in ./src/graph_detailed.png.

The useful Automatic Registration Toolbox we used are listed below:

• acpcdetect v2.1 Linux
• ATRA v1.0 Linux

They are downloaded from NITRC and put in ./utils.

• quqixun/BrainPrep
• nipype
• nilearn
• neurodocker

1. Install neurodocker

   pip install neurodocker

2. Generate Dockerfile using neurodocker

   neurodocker generate docker \
       --pkg-manager apt \
       --base-image neurodebian:bullseye \
       --fsl version=6.0.5.1 \
       --ants version=2.4.1 \
       --miniconda version=latest conda_install="nipype" \
   > Dockerfile.neuro

3. Build the docker image with the generated Dockerfile

   docker build --tag neuro:latest --file Dockerfile.neuro .

Nat Lee