LONDN-MRI

Code for testing and reproducing results for MRI local study project:

Modl and BLIPS code according to the following paper

Anish Lahiri, Guanhua Wang, Sai Ravishankar, Jeffrey A. Fessler, (2021). "Blind Primed Supervised (BLIPS) Learning for MR Image Reconstruction." IEEE Transactions on Medical Imaging http://doi.org/10.1109/TMI.2021.3093770; arXiv preprint arXiv:2104.05028.

The code is made up of three components:

1. single coil two channel study2
2. multi coil two channel global study
3. multi coil two channel local model (with PyTorch > 1.7.0).

Additionally, we used BART to generate the dataset.

MRI local learning

S. Liang, A. Lahiri and S. Ravishankar, "Adaptive Local Neighborhood-based Neural Networks for MR Image Reconstruction from Undersampled Data," in IEEE Transactions on Computational Imaging, doi: 10.1109/TCI.2024.3394770. https://ieeexplore.ieee.org/abstract/document/10510040

Directory overview (under multi_coil_LONDN/)

• make_two_channel_dataset.py specifies and shows how to make the two channel dataset based on the modification from https://github.com/JeffFessler/BLIPSrecon
• global_network_dataset.py specifies the data loader for MRI image loading from multicoil MR measurements for global case.
• local_network_dataset.py and local_network_dataset_oracle.py specifies the data loader for MRI image loading from multicoil MR measurements for noise local case and oracle local case.
• train_local_unet.py can be used for local model training and testing reconstruction from undersampled mulit-coil k-space measurements using UNet training.
• transfer_learning_local_network.py can be used for local model training and testing reconstruction from undersampled mulit-coil k-space measurements using DIDN training.

---

Data avaliable on Dropbox. Just put the data on the your own dataset on the Kspace_data_name= '/mnt/DataA/NEW_KSPACE' and make the image space image based on the kspace data.