AeSPa : Attention-guided Self-supervised Parallel imaging for MRI Reconstruction

Repository for AeSPa : Attention-guided Self-supervised Parallel imaging for MRI Reconstruction accepted in CVPR 2025

Jinho Joo¹*, Hyeseong Kim^1,2*, Hyeyeon Won¹, Deukhee Lee², Taejoon Eo¹**, Dosik Hwang^1,2**

¹Yonsei University, ²Korea Institute of Science and Technology

* denotes equal contribution, ** denotes corresponding authors

<p align="center"><img src='img/knee.png' width="85%"></p> <p align="center"><img src='img/brain.png' width="85%"></p>

Our Goal

We propose a novel zero-shot scan-specific self-supervised reconstruction method for magnetic resonance imaging to reduce scan times. We use only undersampled k-space for training and evaluation.

AeSPa Architecture

<p align="center"><img src='img/mainfigure.png' width="85%"></p>

Patch Note

<span style="font-size: 12px;"> [25.06.10] : We release our code!</span>

Citation

@inproceedings{joo2025aespa,
  title={AeSPa : Attention-guided Self-supervised Parallel imaging for MRI Reconstruction},
  author={Joo, Jinho and Kim, Hyeseong and Won, Hyeyeon and Lee, Deukhee and Eo, Taejoon and Hwang, Dosik},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={5217--5226},
  year={2025}
}

Google scholar

Jinho Joo

Hyeseong Kim

Q&A

If you have a question regarding the code, please email at jinho381 AT yonsei DOT ac DOT kr and hyeseongkim At yonsei DOT ac DOT kr.