INR_for_DynamicMRI

Official code for "Spatiotemporal implicit neural representation for unsupervised dynamic MRI reconstruction"

1. Environmental Requirements

To run the reconstruction demo, the following dependencies are required:

• Python 3.10.X (Important)
• PyTorch 2.0.0
• torchkbnufft 1.4.0
• tiny-cuda-nn 1.7
• imageio 2.18.0
• torchvision, tensorboard, h5py, scikit-image, tqdm, numpy, scipy

2. Sample Data

Download the sample data from https://drive.google.com/file/d/1DIdtHcHUDEqx-qL4930-pz9mxCI8OYMR/view?usp=sharing and put it into the root directory

3. Run the Demos

To run the basic reconstruction demo, please use the following code:

python3 main.py -g 0 -s 13 -r -m

To ablate relative L2 loss, please use the following code:

python3 main.py -g 0 -s 13 -m

To ablate the coarse-to-fine strategy, please use the following code:

python3 main.py -g 0 -s 13 -r

To run the interpolation demo, please use the following code:

python3 main_spatial_interp.py -g 0 -s 34 -r -m

or

python3 main_temporal_interp.py -g 0 -s 34 -r -m

The rest of the parameters can be easily changed by adding arguments to the parser. The detailed definitions of the arguments can be found by:

python3 main.py -h