Pediatric Template

Repository for generating pediatric template

Data

data.neuro.polymtl.ca:datasets/philadelphia-pediatric

Instructions to download: https://intranet.neuro.polymtl.ca/computing-resources/data/git-datasets.html#download

Generating T1w and T2w Templates

Step-by-step walk-through on how to generate T1w and T2w pediatric templates from philadelphia-pediatric dataset. All code below should be run in your command-line (bash) on CCDB. Each step begins from your working directory: /PATH/TO/pediatric-template.

1. Get data from git annex

git clone git@data.neuro.polymtl.ca:datasets/philadelphia-pediatric
cd philadelphia-pediatric
git checkout 31aea09ec124e2aebffef28927731ae635db3f4f
git annex get .
cd ..

2. Dependencies

Install SCT

git clone git@github.com:spinalcordtoolbox/spinalcordtoolbox.git
cd spinalcordtoolbox
git checkout -b 49a40673e6d1521eb7c2d1d6d7b338ab6811448d
./install_sct
conda activate python/envs/venv_sct/
cd ..

Install template pipeline (https://github.com/neuropoly/template)

git clone git@github.com:neuropoly/template.git
cd template
git checkout a7915f4ccfa075a5d31f4ea84bb9761d42710e9e
cd ..

Note

The T1w template and the T2w template should be in the same template space.

3. T1w data preprocessing and normalization

1. cd template
2. cp ../configuration_T1w.json configuration.json
3. python preprocess_normalize.py configuration.json

# Rename derivatives such that the T2w data normalization does not override the outputs of the T1w data normalization
4. mv ../philadelphia-pediatric-processing/derivatives/sct_straighten_spinalcord ../philadelphia-pediatric-processing/derivatives/sct_straighten_spinalcord_T1w
5. mv ../philadelphia-pediatric-processing/derivatives/template ../philadelphia-pediatric-processing/derivatives/template_T1w

4. T2w data preprocessing and normalization

Once the template space for using the T1w data, a new template space is not needed to be created using the T2w images. This is to ensure that both the templates are in the same template space. For this, follow the steps below:

1. Comment out the `generate_initial_template_space` method in the preprocess_normalize.py script
2. cp ../configuration_T2w.json configuration.json
3. python preprocess_normalize.py configuration.json

# Rename derivatives
4. mv ../philadelphia-pediatric-processing/derivatives/sct_straighten_spinalcord ../philadelphia-pediatric-processing/derivatives/sct_straighten_spinalcord_T2w
5. mv ../philadelphia-pediatric-processing/derivatives/template ../philadelphia-pediatric-processing/derivatives/template_T2w

5. Log in to Digital Research Alliance of Canada (the Alliance) High-Performance Computer (HCP)

Refer to NeuroPoly Lab Manual for further specifications.

6. Set up correct environment on Alliance HPC

cd PATH/TO/scratch
mkdir pediatric-template
cd pediatric-template
module load StdEnv/2020  gcc/9.3.0 minc-toolkit/1.9.18.1 python/3.8.10
pip install --upgrade pip
pip install scoop
git clone https://github.com/vfonov/nist_mni_pipelines.git
nano ~/.bashrc
    :'
    export PYTHONPATH="${PYTHONPATH}:/PATH/TO/scratch/philadelphia-pediatric/derivatives/model_nl_all_T2w/code/nist_mni_pipelines"
    export PYTHONPATH="${PYTHONPATH}:/PATH/TO/scratch/philadelphia-pediatric/derivatives/model_nl_all_T2w/code/nist_mni_pipelines/"
    export PYTHONPATH="${PYTHONPATH}:/PATH/TO/scratch/philadelphia-pediatric/derivatives/model_nl_all_T2w/code/nist_mni_pipelines/ipl/"
    export PYTHONPATH="${PYTHONPATH}:/PATH/TO/scratch/philadelphia-pediatric/derivatives/model_nl_all_T2w/code/nist_mni_pipelines/ipl"
    '
source ~/.bashrc
pip install "git+https://github.com/NIST-MNI/minc2-simple.git@develop_new_build#subdirectory=python"

7. Bring necessary data to Alliance HCP using rsync

The data contained in philadelphia-pediatric-processing/derivatives/template_T1w and philadelphia-pediatric-processing/derivatives/template_T2w
should be on your Alliance HCP scratch directory. Your directory PATH/TO/SCRATCH/template-pediatric should now have the following structure:

nist_mni_pipelines
my_job.sh
make_subjects_csv.sh
generate_template.py
template_T1w/
    sub*T1w_straight_norm.mnc
    template_mask.mnc
template_T2w/
    sub*T2w_straight_norm.mnc
    template_mask.mnc

8. Bring necessary data to Alliance HCP using rsync

The data contained in philadelphia-pediatric-processing/derivatives/template_T1w and philadelphia-pediatric-processing/derivatives/template_T2w
should be on your Alliance HCP scratch directory. Your directory PATH/TO/SCRATCH/template-pediatric should now have the following structure:

nist_mni_pipelines
my_job.sh
make_subjects_csv.sh
generate_template.py
template_T1w/
    sub*T1w_straight_norm.mnc
    template_mask.mnc
template_T2w/
    sub*T2w_straight_norm.mnc
    template_mask.mnc

9. Create T1w template

The template-generation process is an iterative averaging process.

• It does a voxel-wise averaging of sub-*_T2w_straight_norm.mnc in template_mask.mnc space to create avg.001.mnc
• This process is then repeated in avg.001.mnc, to generate avg.002.mnc.
• etc.

cd PATH/TO/SCRATCH/template-pediatric
# generate subjects.csv (list of full path to `sub-*_T1w_straight_norm.mnc` files and `template_mask.mnc`, used by `generate_template.py`)
./make_subjects_csv.sh template_T1w  
module load StdEnv/2020  gcc/9.3.0 minc-toolkit/1.9.18.1 python/3.8.10
sbatch --account=def-jcohen --time=48:00:00 --mem-per-cpu 8000 my_job.sh

Keep running for several iterations. Sometimes, there is not enough time for an iteration to finish completing and the job needs to be resubmitted. Each iteration i generates the following files upon successful completion:

model_nl_all/
    i/
        # 201 .mnc files
        # 112 .xfm files
    avg.i.mnc
    avg.i_mask.mnc
    sd.i.mnc

avg.i.mnc is the template generated by iteration i.

10. Convert to T1w template to NIFTI

After N iterations, if you are satisfied with your template (avg.N.mnc), you can convert it to NIFTI format and save it!

module load StdEnv/2020  gcc/9.3.0 minc-toolkit/1.9.18.1 python/3.8.10
cd PATH/TO/SCRATCH/template-pediatric
mv model_nl_all model_nl_all_T1w # such that T2w template-generation does not use the T1w averages
cd model_nl_all_T1w
mnc2nii avg.N.mnc
gzip avg.N.nii
mv avg.N.nii.gz pediatric-template_T1w.nii.gz

11. Create T2w template

Same as steps 10 and 9, but with T2w data!

cd PATH/TO/SCRATCH/template-pediatric
# generate subjects.csv (list of full path to `sub-*_T2w_straight_norm.mnc` files and `template_mask.mnc`, used by `generate_template.py`)
./make_subjects_csv.sh template_T2w  
module load StdEnv/2020  gcc/9.3.0 minc-toolkit/1.9.18.1 python/3.8.10
sbatch --account=def-jcohen --time=48:00:00 --mem-per-cpu 8000 my_job.sh

When you are satisfied with your template, you can convert it to NIFTI:

module load StdEnv/2020  gcc/9.3.0 minc-toolkit/1.9.18.1 python/3.8.10
mv model_nl_all model_nl_all_T2w
cd model_nl_all_T2w
mnc2nii avg.N.mnc
gzip avg.N.nii
mv avg.N.nii.gz pediatric-template_T2w.nii.gz