USAGE NOTES

Note

• We recommand to run freesurfer first, becuase all files we need are freesurfer output files.

GAN-MAT structure

GAN-MAT
├── docs
├── functions
│   └── model
├── parcellations
├── template
├── folder_setting.ipynb
├── gan-mat
└── README.md

• functions: : A folder containing (i) the GAN model (functions/model) to synthesize the T2-weighted MRI from the T1-weighted MRI, and (ii) necessary scripts to calculate myelin-sensitive proxy (T1w/T2w ratio) and (iii) to generate microstructural profile covariance (MPC) matrix, microstructural gradients, and moment features.

• parcellations: A folder containing 18 different atlases: aparc, aparc-a2009s, economo, glasser-360, schaefer-100~1000, vosdewael-100~400.

• template: A folder containing the MNI 0.8mm T1 template for initial registration.

• folder_setting.ipynb: A file converting input data to an appropriate format to run the pipeline.

• gan-mat: A main script to run GAN-MAT.

Run GAN-MAT

1. Set up the directory with a specific format using folder_setting.ipynb as follows:

input_dir
├── 105923                                        # Subject ID
│   └── T1w
│       └─── 105923                               # Subject ID
│           ├── anat
│           │   └── surfaces
│           │       └── micro_profiles
│           ├── label                             # freesurfer output
│           │   ├── lh.cortex.label
│           │   └── rh.cortex.label
│           ├── mri                               # freesurfer output
│           │   └── orig.mgz
│           └── surf                              # freesurfer output
│               ├── lh.area
│               ├── lh.area.pial
│               ├── lh.pial
│               ├── lh.sphere.reg
│               ├── lh.white
│               ├── rh.area
│               ├── rh.area.pial
│               ├── rh.pial
│               ├── rh.sphere.reg
│               └── rh.white
└── ...

2. Run gan-mat.

gan-mat -input_dir /INPUT/DATA/PATH -output_dir /OUTPUT/PATH <Options>

Options	Description
batch_size <num>	Number of the batch size while synthesizing the T2-weighted MRI (default = 1)
threads <num>	Number of threads (default = 6)
T2	Synthesize T2-weight MRI and terminate
myelin	Compute myelin-sensitive proxy (T1w/T2w ratio) and terminate
matrix	Calculate microstructural profile covariance (MPC) matrix and terminate
gradients	Generate microstructural gradient and terminate

Note

• If options of features (T2, myelin, matrix, gradients) are not provided, the pipeline will generate all outputs.

• All the outputs will be stored in the individual subject’s folder.

Warning

The 10GB of GPU memory is required. In the case of low GPU capacity, change the device variable in the ~/GAN-MAT/functions/model/t2.py.