Jing Cheng1, Wenqi Huang1, Zhuoxu Cui1, Ziwen Ke1, Leslie Ying2, Haifeng Wang1, Yanjie Zhu1, and Dong Liang1
1Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, 2University at Buffalo, The State University of New York, Buffalo, Buffalo, NY, United States
1Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, 2University at Buffalo, The State University of New York, Buffalo, Buffalo, NY, United States
We propose a new DL-based approach that implicitly
learns the data consistency, which is corresponding to the real distribution of system noise. The DC term and
the prior knowledge are both embedded in the weights of the networks to provides an utterly implicit learning of reconstruction
model.
Figure 2. Reconstructions in spatial domain
under different acceleration factors with the zoom-in images of the enclosed
part and the corresponding error maps.
Figure 1. Illustration of the proposed
Learned DC. The architecture of each iteration is corresponding to Eq. (6),
where the k-space sub-network has 2 layers and image domain sub-network has 3
layers. The numbers on the layers indicate the number of filters in that
convolutional layer.
