Unsupervised MRI Reconstruction with Generative Adversarial Networks

• 2020 Elizabeth Cole, Stanford University (ekcole@stanford.edu)

Make sure the python requirements are installed

pip3 install -r requirements.txt

The setup assumes that the latest Berkeley Advanced Reconstruction Toolbox is installed [1]. The scripts have all been tested with v0.4.01.

We will first download data, generate sampling masks, and generate TFRecords for training. The datasets downloaded are fully sampled volumetric knee scans from mridata [2]. The setup script uses the BART binary. In a new folder, run the follwing script:

python3 mri_util/setup_mri.py -v

The training of the unsupervised GAN can be ran using the following script:

python3 train_unsupervised.py dataset_dir model_dir

where dataset_dir is the folder where the knee datasets were saved to, and model_dir will be the top directory where the models will be saved to.

Testing can be ran using:

python3 test_unsupervised.py dataset_dir model_dir

The training of the supervised GAN can be ran using the following script:

python3 train_supervised.py dataset_dir model_dir

where dataset_dir is the folder where the knee datasets were saved to, and model_dir will be the top directory where the models will be saved to.

Testing can be ran using:

python3 test_supervised.py dataset_dir model_dir

For any issues or questions, please open an issue on the github repo or contact Elizabeth at ekcole@stanford.edu.