Disclaimer: deepmriprep is not related to fMRIPrep or sMRIPrep and is not part of the NiPreps framework

This repo contains scripts to train neural networks used in deepmriprep

Installation 🛠️

Install CAT12 (version 12.8.2 was used in the publication)

To install pycairo (dependency of niftiai) run

sudo apt install libcairo2-dev pkg-config python3-dev

torch with (your version of) CUDA support should be installed first via the proper install command

Then use requirements.txt to install the remaining dependencies

Download MRIs 📥

Pick a folder on a fast disk(=SSD) on your system with ~500GB of free space (per default data)

If that folder should not be data (the default),

1. copy and paste the data folder into your desired folder
2. adapt data_path (and path, see comment) at the beginning of each of the 7 scripts

Download the T1w MRIs listed in data/csvs/openneuro_hd.csv from OpenNeuro

Preprocess

The downloaded MRIs should be placed in data/t1 with the filenames from openneuro_hd.csv

Applying CAT12 to these MRIs should—e.g., for the p0 output of filename 0009_sub-06—result in

data/t1/CAT12.8.2/mri/p00009_sub-06.nii

with this filepath-pattern applied to all 685 filenames and CAT12 output modalities.

In 2_prep_segment.py, a rerun of CAT12 on the img_05mm files should result in e.g.

data/img_05mm/CAT12.8.2/mri/p00009_sub-06.nii

Run scripts

Run the 7 scripts (+read the comments) 1_prep_warp.py-7_compile_models.py!

The trained models (e.g. the warp model) can be directly plugged into deepmriprep like this:

from deepmriprep import run_preprocess

run_preprocess(bids_dir='path/to/bids', warp_model_path='path/to/warp_model.pt')