/Synb0-DISCO

Primary LanguageMATLAB

Synb0-DISCO (TIGRLab)

Synb0-DISCO is a tool used for generating synthetic b0s using deep learning using only a T1. This is useful when doing diffusion imaging without fieldmap/phasediff files. Find more about this tool at the original repo or the paper.

The TIGRLab version of the repo has modifications to make it easier to run on our system.

Installation

Pull this repo.

git clone https://github.com/TIGRLab/Synb0-DISCO

Pull the pix2pix repo to Synb0-DISCO/src/python. To be safe, switch to revision 6d9e17390a4b2e186bc45427c37f4ff42409bb40 which is confirmed to work.

git clone https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix Synb0-DISCO/src/python/pytorch-CycleGAN-and-pix2pix/
git -C Synb0-DISCO/src/python/pytorch-CycleGAN-and-pix2pix/ checkout 6d9e1739

Copy the training data from smansour's projects folder.

mkdir -p Synb0-DISCO/src/checkpoints
cp -r /projects/smansour/Synb0_DISCO_checkpoints/* Synb0-DISCO/src/checkpoints/

Load FSL 6.0.1, matlab R2017b.

module load FSL/6.0.1 matlab/R2017b

Create a python (3.5.5) environment with the dependencies from Synb0-DISCO/src/requirements.txt.

*note, install python environments from here: https://github.com/pyenv/pyenv, then install the software for virtual environments from here https://github.com/pyenv/pyenv-virtualenv

Note: Always activate your python environment after loading the modules from the previous step. This is because loading those modules sometimes override the python environments.

pyenv virtualenv 3.5.5 synb0
pyenv activate synb0
pip install -r Synb0-DISCO/src/requirements.txt

Congrats! You are now ready to generate synthetic b0s.

Usage

Note, prior to running this command, make sure your T1 images are oriented correctly (fslreorient2std) and de-biased N4BiasFieldCorrection - otherwise the image may be cut off

In the following commands, <synb0_root> should always point to the folder where you pulled the repo, Synb0-DISCO. This is required to import all the scripts in it and the training data.

To generate a b0 given a single T1:

python Synb0-DISCO/src/gen_synb0_single.py <t1_file> <synb0_root> <output_folder>
#note your T1 image cannot have an extension (e.g., if it ends in .nii.gz the script will not run)

first install pybids

pip install pybids

To generate a b0 given a bids directory:

python Synb0-DISCO/src/gen_synb0_bids.py <bids_dir> <synb0_root> <output_folder>