Skip to content

Latest commit

 

History

History
35 lines (22 loc) · 1.57 KB

cardiac-mr-motion-artefact-correction-from-k-space-using-deep-learning-based-reconstruction-2018-200116.md

File metadata and controls

35 lines (22 loc) · 1.57 KB

Paper:

Cardiac MR Motion Artefact Correction from K-space using Deep Learning-based Reconstruction

Author:

Ilkay Oksuz, James Clough, Aurelien Bustin, Gastao Cruz, Claudia Prieto, Rene Botnar, Daniel Rueckert, Julia A. Schnabel, Andrew P. King

Year:

2018

Notes:

阅读日期:2020.01.16

泛读。这篇文章研究的是如何去除 cardiac MRI 图像的 motion 伪影。使用的方法是 kspace to image 的 GAN。方法很简单,没有什么参考价值。网络结构很简单,先是两个全连接层,然后用卷积和反卷积,最后是一个 1x1 卷积层。判别网络也很简单。网络结构示意图:

训练时,作者先用 ImageNet 的kspace 做预训练,然后再用 MRI 数据训练。另外作者提到了一个 regularizer,但是没有说明具体内容。具有 motion 伪影的数据的产生方法是:假设是笛卡尔采样。对 cartesian sampling lines 换成其他的 cardiac phase 的sampling lines

Bibtex:

@inproceedings{oksuz2018cardiac,
  title={Cardiac mr motion artefact correction from k-space using deep learning-based reconstruction},
  author={Oksuz, Ilkay and Clough, James and Bustin, Aurelien and Cruz, Gastao and Prieto, Claudia and Botnar, Rene and Rueckert, Daniel and Schnabel, Julia A and King, Andrew P},
  booktitle={International Workshop on Machine Learning for Medical Image Reconstruction},
  pages={21--29},
  year={2018},
  organization={Springer}
}