The project repository for SC-GAN: 3D self-attention conditional GAN with spectral normalization for multi-modal neuroimaging synthesis

The data directory should be organized as the following structure:

data
│
└───train
|   |
│   └─── subject 1
|   |         modality_1.nii.gz
|   |         modality_2.nii.gz
|   |         ...
|   |         target.nii.gz
│   └─── subject 2
|   |         modality_1.nii.gz
|   |         modality_2.nii.gz
|   |         ...
|   |         target.nii.gz
│   ...
|
└───val
|   |
│   └─── subject A
|   |         modality_1.nii.gz
|   |         modality_2.nii.gz
|   |         ...
|   |         target.nii.gz
│   └─── subject B
|   |         modality_1.nii.gz
|   |         modality_2.nii.gz
|   |         ...
|   |         target.nii.gz
│   ...
|   
└───test
    |
    └─── subject a
    |         modality_1.nii.gz
    |         modality_2.nii.gz
    |         ...
    └─── subject b
    |         modality_1.nii.gz
    |         modality_2.nii.gz
    |         ...
    ...

To evaluate SC-GAN, we used ADNI data. To access ADNI data, please visit: https://ida.loni.usc.edu/

python 3 is required and python 3.6.9 was used in the study.

Requirements can be found at requirement.txt.

Please use pip install requirement.txt to install the requirements

SC-GAN was assessed for positron emission tomography (PET), Fractional anisotropy (FA) and mean diffusivity (MD) synthesis from tasks using T1w and FLAIR magnetic resonance images (MRI).

Training script is at ./training

Use the following command to run the training script:

python training.py --trainig_size= --gpu= --epoches= --img_size= --data_dir= --modalities= --logdir=

Add more parameters configuration to do hyperparameter tuning:

Testing script is at ./testing

Use the following command to run the testing script:

python testing.py --test_size= --gpu= --img_size= --data_dir= --modalities= --logdir= --output=

If you use this code for your research, please be familiar with the LICENSE and cite our paper.

Haoyu Lan
INI Microstructural imaging Group (IMG)
USC Stevens Neuroimaging and Informatics Institute
Keck School of Medicine, University of Southern California

Lan H, Toga AW, Sepehrband F. Three- dimensional self- attention conditional GAN with spectral normalization for multimodal neuroimaging synthesis. Magn Reson Med. 2021;00:1– 16. https://doi.org/10.1002/mrm.28819
[bioRxiv copy]