多尺度残差学习结合Dilformer的双流医学图像配准网络

彭静,闫佳荣,刘佳英,魏子易,白珊,邓亚红

Multi-scale residual learning combined with Dilformer for dual-stream medical image registration network

Jing PENG,Jiarong YAN,Jiaying LIU,Ziyi WEI,Shan BAI,Yahong DENG

表 1 各医学图像配准模型在IXI数据集上的定量分析结果

Tab.1 Quantitative analysis results of each medical image registration model on IXI dataset

模型 DSC/% HD95 $ {R}_{{\mathrm{Jac}}} $ Np/106 Flops/109 t/s SyN 64.1±1.6 10.878 <0.001 — — 41.83 LDDMM 67.9±1.5 10.672 <0.001 — — 31.60 VoxelMorph 72.9±1.2 10.431 1.604 0.28 304.31 0.21 CycleMorph 73.6±1.8 10.348 1.523 0.36 160.57 0.25 LKU-Net 76.4±1.1 10.107 0.713 2.09 272.63 0.23 ViT-V-Net 73.0±1.2 10.305 1.681 31.56 389.21 0.31 TransMorph 75.1±1.3 9.741 1.565 107.76 713.54 0.38 TransMatch 75.0±3.0 9.653 1.594 70.71 717.23 0.37 PIVit 75.7±2.6 9.327 0.454 0.65 71.19 0.26 RDP-Net 76.1±1.1 9.156 0.136 8.92 2889.00 0.27 MSRD-Net 76.9±2.0 8.937 0.029 13.95 863.17 0.29