IEEE Transactions on Medical Imaging 2021

CINE_MRI_CONVLSTM

Cine Cardiac MRI Motion Artifact Reduction Using a Recurrent Neural Network

Qing Lyu, Hongming Shan, Yibin Xie, Alan C. Kwan, Yuka Otaki, Keiichiro Kuronuma, Debiao Li, Ge Wang

Current or former lab member. Select a name to see all of their projects.

Method overview of CINE_MRI_CONVLSTM
Architecture of the proposed recurrent network, with two ConvLSTM branches and an encoder-decoder sub-network (Fig. 2 of the paper).

Abstract

Cine cardiac MRI is widely used for diagnosis, but its long scan time leads to motion artifacts and patient discomfort. This work proposes a recurrent generative adversarial network to reduce those artifacts. Bi-directional ConvLSTMs capture long-range temporal features across frames, and multi-scale convolutions gather local and global features.

Experiments show better image quality than existing state-of-the-art methods. The model can also generate missing intermediate frames from adjacent frames, which improves temporal resolution.

What’s in the repository

  • The network and training code
  • An SSIM implementation used by the loss

Citation

@article{lyu2021cine,
  title   = {Cine Cardiac MRI Motion Artifact Reduction Using a Recurrent Neural Network},
  author  = {Lyu, Qing and Shan, Hongming and Xie, Yibin and Kwan, Alan C. and Otaki, Yuka and Kuronuma, Keiichiro and Li, Debiao and Wang, Ge},
  journal = {IEEE Transactions on Medical Imaging},
  volume  = {40},
  number  = {8},
  pages   = {2170--2181},
  year    = {2021},
  doi     = {10.1109/TMI.2021.3073381}
}
Back to projects