Skip to content

ReconFormer: Accelerated MRI Reconstruction Using Recurrent Transformer

License

Notifications You must be signed in to change notification settings

guopengf/ReconFormer

Repository files navigation

ReconFormer

ReconFormer: Accelerated MRI Reconstruction Using Recurrent Transformer

Pytorch Code for the paper "ReconFormer: Accelerated MRI Reconstruction Using Recurrent Transformer"

Updates:

🚀 : We release training and testing code
🚀 : We release pre-trained weights for fastMRI

Requirements

python=3.6
pytorch=1.7.0

Please refer conda_environment.yml for more dependencies.

Inroduction

Accelerating magnetic resonance imaging (MRI) reconstruction process is a challenging ill-posed inverse problem due to the excessive under-sampling operation in k-space. In this paper, we propose a recurrent transformer model, namely ReconFormer, for MRI reconstruction which can iteratively reconstruct high fidelity magnetic resonance images from highly under-sampled k-space data. In particular, the proposed architecture is built upon Recurrent Pyramid Transformer Layers (RPTLs), which jointly exploits intrinsic multi-scale information at every architecture unit as well as the dependencies of the deep feature correlation through recurrent states. Moreover, the proposed ReconFormer is lightweight since it employs the recurrent structure for its parameter efficiency.

Dataset Preparation

Prepare the dataset in the following structure for easy use of the code.The provided data loaders is ready for this this format and you may change it as your need.

                   |-- 
                   |                       
                   |                |--xxx.h5  
Dataset Folder-----|      |--train--|...
                   |      |         |...
                   |      |                  
                   |      |         
                   |--PD -|
                          |
                          |         |--xxx.h5 
                          |-- val --|...  
                                    |...

Links for downloading the public datasets:

  1. fastMRI Dataset - Link
  2. HPKS Dataset - We don't obtain the permission from Johns Hopkins Hospital to release this dataset.

Preprocessed fastMRI (OneDrive) - Link
Password: pguo4
Note: In preprocessed fastMRI, We didn't modify the original fastMRI data and just make the format compatible with our DataLoader.

Run

Clone this repo

git clone [email protected]:guopengf/ReconFormer.git

Set up conda environment

cd ReconFormer
conda env create -f conda_environment.yml
conda activate recon

Train ReconFormer

bash run_recon_exp.sh

Monitor the traning process

tensorboard --logdir 'Dir path for saving checkpoints'

Test

(Download pre-trained weights Password: [email protected])

bash run_recon_eval.sh

Ackonwledgements

We give acknowledgements to fastMRI, Swin-Transformer , and SwinIR.

Citation

@article{guo2022reconformer,
  title={ReconFormer: Accelerated MRI Reconstruction Using Recurrent Transformer},
  author={Guo, Pengfei and Mei, Yiqun and Zhou, Jinyuan and Jiang, Shanshan and Patel, Vishal M},
  journal={arXiv preprint arXiv:2201.09376},
  year={2022}
}

About

ReconFormer: Accelerated MRI Reconstruction Using Recurrent Transformer

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published