Ziqi Yu1, Yuting Zhai1, Wenjing Xu1, Xiang Chen1, and Xiao-Yong Zhang1
1Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University, Shanghai, China
1Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University, Shanghai, China
We present the U-Net with Nonlocal Position-aware (NPA) block using domain knowledge transfer. The results showd that our end-to-end method achieves high dice scores in several MR modalities with ultrafast processing speed which is two orders of magnitude faster than atlas-based methods.
Figure 2: Structure of the network. Nonlocal position-aware block (NPAB) is deployed in the bottleneck of U-Net. The number of channels indicates below convolution blocks. The parameters are shared and reused in two training stages.
Figure 4: Comparison of different methods performance in same-domain(T2) and cross-domain(SWI and ASL).