Optics and Precision Engineering, Volume. 30, Issue 10, 1203(2022)

Image registration based on residual mixed attention and multi-resolution constraints

Mingna ZHANG1, Xiaoqi LÜ1,2、*, and Yu GU1
Author Affiliations
  • 1Key Laboratory of Rattern Recognition and Intelligent Image Processing, School of Information Engineering,Inner Mongolia University of Science and Technology, Baotou0400, China
  • 2School of Information Engineering, Inner Mongolia University of Technology, Hohhot010051, China
  • show less
    Figures & Tables(18)
    Diagram of registration network model
    Architecture of MAMReg-Net network
    Diagram of residual mixed attention
    Structure of Mask branching
    Non-Local module
    Images before and after preprocessing
    Image of 12 examples of anatomical structures
    Samples registration result from test images
    Color overlay images before and after registration
    Registration results using different methods
    Histogram of the average Dice score of the 12 anatomical structures on the test images
    Histogram of the average ASD score of the 12 anatomical structures on the test images
    Dice score boxplot of 12 anatomical structures in ablation experiment
    ASD score boxplot of 12 anatomical structures in ablation experiment
    • Table 1. Registration accuracy of different methods

      View table
      View in Article

      Table 1. Registration accuracy of different methods

      MethodAverage DiceAverage ASD
      Affine0.614±0.0711.673±0.369
      SYN280.783±0.0480.866±0.142
      VoxelMorph170.804±0.0290.804±0.137
      MAMReg-Net0.817±0.0350.789±0.205
    • Table 2. Comparison of registration time between different methods

      View table
      View in Article

      Table 2. Comparison of registration time between different methods

      Method

      Average registration

      time/s

      SYN2835
      Tensorflow-VoxelMorph170.74
      Pytorch-VoxelMorph0.16
      MAMReg-Net0.34
    • Table 3. Comparison of registration accuracy of ABIDE dataset

      View table
      View in Article

      Table 3. Comparison of registration accuracy of ABIDE dataset

      MethodAverage DiceAverage ASD
      Affine0.540±0.0351.951±0.281
      SYN280.740±0.0131.085±0.092
      VoxelMorph170.726±0.0201.179±0.115
      MAMReg-Net0.734±0.0251.162±0.163
    • Table 4. Comparison of registration accuracy of ablation experiments

      View table
      View in Article

      Table 4. Comparison of registration accuracy of ablation experiments

      MethodDiceASD
      BaseMulti-resolution-constraintResidual-attentionNon-local
      0.807±0.0380.836±0.211
      0.809±0.0360.819±0.200
      0.813±0.0380.805±0.213
      0.815±0.0380.800±0.217
      0.817±0.0350.789±0.205
    Tools

    Get Citation

    Copy Citation Text

    Mingna ZHANG, Xiaoqi LÜ, Yu GU. Image registration based on residual mixed attention and multi-resolution constraints[J]. Optics and Precision Engineering, 2022, 30(10): 1203

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Information Sciences

    Received: Dec. 22, 2021

    Accepted: --

    Published Online: Jun. 1, 2022

    The Author Email: LÜ Xiaoqi (lxiaoqi@imust.edu.cn)

    DOI:10.37188/OPE.20223010.1203

    Topics