Pytorch implementation of RAKI paper with mild changes and optimizations [1]
Clone the Repo:
git clone https://github.com/geopi1/DeepMRI.git
Download the Datasets:
In this site select any of the available MRI scans and download to a folder
- An auto download script will be added soon
- Setup conda
This will create a working environment named DeepMRI
conda env create -f env.yml
- Setup can also be performed with pip (virtual env) via the requirements.txt file
python3 -m venv DeepMRI pip install -r requirements.txt
- Run save_raw_data_to_pickle.py to save the .h5 files from mridata.org as a pickle with np matrices
or
python save_raw_data_to_pickle.py -p [path_to_wanted_folder]
python save_raw_data_to_pickle.py --data_path [path_to_wanted_folder]
All the hyperparameters of the code are saved as .json in config.json. Please look at the number of epochs.
To run the code, activate the conda environment
conda activate DeepMRI
or select the appropriate python interpreter path and run:
python main.py
a specific path can be added to the command line (instead of the config file)
python main.py -data /path/to/data/folder
Each run records the training process, saving the learing_rate (lr) and the loss. To view these live via tensorboard:
- Navigate to the appropriate folder
- Open command line or terminal
- from the proper conda env (or it tensorboard is in path) type:
tensorboard --logdir logs_dir/
This project is licensed under the MIT License - see the LICENSE.md file for details
[1] Akçakaya, Mehmet et al. “Scan-specific robust artificial-neural-networks for k-space interpolation (RAKI) reconstruction: Database-free deep learning for fast imaging.” Magnetic resonance in medicine vol. 81,1 (2019): 439-453. doi:10.1002/mrm.27420
[2] mridata.org