Non-rigid Motion Correction for MRI Reconstruction via Coarse-To-Fine Diffusion Models

by Frederic Wang and Jonathan I. Tamir

Paper link

Dependencies

• torch
• kornia
• numpy
• scipy
• matplotlib
• sigpy
• bart (see https://github.com/mrirecon/bart for installation details)

File Descriptions

CMR_data/

Sample cardiac cine SAX data of a single patient from CMRxRecon 2024 dataset.

brain_data/

Sample axial T1 post contrast data from fastMRI.

CardiacMotion.ipynb

Notebook containing code used to reconstruct motion-corrupted cine cardiac MRI from CMRxRecon using C2F (and benchmarks).

PreprocessData.ipynb

Notebook containing code used to generate test data in the format of folders CMR_data and brain_data

SimulatedMotion.ipynb

Notebook containing code used to simulate rigid and nonrigid motion and jointly reconstruct and perform motion correction using C2F (and benchmarks).

utils.py

Utility functions for diffusion model sampling, deforming images, and learning deformation fields.

References

@article{wang2025non,
  title={Non-rigid Motion Correction for MRI Reconstruction via Coarse-To-Fine Diffusion Models},
  author={Wang, Frederic and Tamir, Jonathan I},
  journal={arXiv preprint arXiv:2505.15057},
  year={2025}
}

We also use code and data from the following sources:

@inproceedings{Karras2022edm,
  author    = {Tero Karras and Miika Aittala and Timo Aila and Samuli Laine},
  title     = {Elucidating the Design Space of Diffusion-Based Generative Models},
  booktitle = {Proc. NeurIPS},
  year      = {2022}
}

@article{wang2024cmrxrecon,
  title={CMRxRecon: A publicly available k-space dataset and benchmark to advance deep learning for cardiac MRI},
  author={Wang, Chengyan and Lyu, Jun and Wang, Shuo and Qin, Chen and Guo, Kunyuan and Zhang, Xinyu and Yu, Xiaotong and Li, Yan and Wang, Fanwen and Jin, Jianhua and others},
  journal={Scientific Data},
  volume={11},
  number={1},
  pages={687},
  year={2024},
  publisher={Nature Publishing Group UK London}
}

@article{knoll2020fastmri,
  title={fastMRI: A publicly available raw k-space and DICOM dataset of knee images for accelerated MR image reconstruction using machine learning},
  author={Knoll, Florian and Zbontar, Jure and Sriram, Anuroop and Muckley, Matthew J and Bruno, Mary and Defazio, Aaron and Parente, Marc and Geras, Krzysztof J and Katsnelson, Joe and Chandarana, Hersh and others},
  journal={Radiology: Artificial Intelligence},
  volume={2},
  number={1},
  pages={e190007},
  year={2020},
  publisher={Radiological Society of North America}
}