Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Acta Optica Sinica, Volume. 44, Issue 19, 1915002(2024)

Design of Large Deformable Lung Image Registration Network with Adaptive Window

Jianbing Yi1,2、*, Xi Chen1,2, Feng Cao1,2, Shuxin Yang1,2、**, and Xin Chen1,2
Author Affiliations
  • 1School of Information Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, Jiangxi , China
  • 2Jiangxi Provincial Key Laboratory of Multidimensional Intelligent Perception and Control, Ganzhou 341000, Jiangxi , China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (18)
    Equations (0)
    References (29)
    Cited By (2)
    Tools
    Paper Information
    Topics
    Figures & Tables(18)
    Example of edge convolution operation

    Fig. 1. Example of edge convolution operation

    Download full size
    Flowchart of registration method

    Fig. 2. Flowchart of registration method

    Download full size
    Feature correlation coefficient calculation module based adaptive window

    Fig. 3. Feature correlation coefficient calculation module based adaptive window

    Download full size
    U-Net network based on multiscale dense connection fusion

    Fig. 4. U-Net network based on multiscale dense connection fusion

    Download full size
    MSRN. (a) Multi-scale residual convolution; (b) multi-scale convolution network

    Fig. 5. MSRN. (a) Multi-scale residual convolution; (b) multi-scale convolution network

    Download full size
    DGCN structure

    Fig. 6. DGCN structure

    Download full size
    CT images before and after preprocessing. (a) Original image; (b) preprocessed image

    Fig. 7. CT images before and after preprocessing. (a) Original image; (b) preprocessed image

    Download full size
    Boxplot and scatterplot of TRE for 10 cases on DIR-lab dataset. (a) Boxplot of TRE from maximum exhalation phase to maximum inhalation phase on DIR-lab dataset; (b) scatterplot of TRE before and after registration on DIR-lab dataset

    Fig. 8. Boxplot and scatterplot of TRE for 10 cases on DIR-lab dataset. (a) Boxplot of TRE from maximum exhalation phase to maximum inhalation phase on DIR-lab dataset; (b) scatterplot of TRE before and after registration on DIR-lab dataset

    Download full size
    Boxplot and scatterplot of TRE for 10 cases on COPDgene dataset. (a) Boxplot of TRE from maximum exhalation phase to maximum inhalation phase on COPDgene dataset; (b) scatterplot of TRE before and after registration on COPDgene dataset

    Fig. 9. Boxplot and scatterplot of TRE for 10 cases on COPDgene dataset. (a) Boxplot of TRE from maximum exhalation phase to maximum inhalation phase on COPDgene dataset; (b) scatterplot of TRE before and after registration on COPDgene dataset

    Download full size
    Boxplot and scatterplot of TRE for 6 cases on Creatis dataset. (a) Boxplot of TRE from maximum exhalation phase to maximum inhalation phase on Creatis dataset; (b) scatterplot of TRE before and after registration on Creatis dataset

    Fig. 10. Boxplot and scatterplot of TRE for 6 cases on Creatis dataset. (a) Boxplot of TRE from maximum exhalation phase to maximum inhalation phase on Creatis dataset; (b) scatterplot of TRE before and after registration on Creatis dataset

    Download full size
    Effect of dense graph convolution module on number of TRE outliers in COPDgene dataset

    Fig. 11. Effect of dense graph convolution module on number of TRE outliers in COPDgene dataset

    Download full size
    Registration results for case1 on DIR-lab dataset. (a)‒(c) Fixed images; (d)‒(f) moving images; (g)‒(i) deformation field color-mapped visualizations; (j)‒(l) deformation field grid visualizations; (m)‒(o) warped images; (p)‒(r) heat maps of intensity difference between moving image and fixed image; (s)‒(u) heat maps of intensity difference between warped image and fixed image

    Fig. 12. Registration results for case1 on DIR-lab dataset. (a)‒(c) Fixed images; (d)‒(f) moving images; (g)‒(i) deformation field color-mapped visualizations; (j)‒(l) deformation field grid visualizations; (m)‒(o) warped images; (p)‒(r) heat maps of intensity difference between moving image and fixed image; (s)‒(u) heat maps of intensity difference between warped image and fixed image

    Download full size
    Registration results for different algorithms. (a)‒(c) Moving images; (d)‒(f) fixed images; (g)‒(i) registration results of Graphregnet algorithm; (j)‒(l) registration results of our method

    Fig. 13. Registration results for different algorithms. (a)‒(c) Moving images; (d)‒(f) fixed images; (g)‒(i) registration results of Graphregnet algorithm; (j)‒(l) registration results of our method

    Download full size

    • Table 1. TRE for different algorithms on DIR-lab dataset

      View table

      Table 1. TRE for different algorithms on DIR-lab dataset

      CaseInitialGraphregnetLRNDuanHPRNOurs
      Mean8.46 （6.58）1.49 （—）1.78 （1.56）1.59 （—）2.04 （1.42）1.21 （0.83）
      13.89 （2.78）0.86 （—）1.14 （0.71）1.05 （—）1.22 （0.71）0.98 （0.52）
      24.34 （3.90）0.90 （—）1.04 （1.13）0.87 （—）1.24 （0.76）0.94 （0.47）
      36.94 （4.05）1.13 （—）1.44 （0.84）1.18 （—）1.49 （0.61）1.17 （0.66）
      49.83 （4.86）1.61 （—）1.54 （1.20）1.71 （—）1.70 （0.68）1.38 （0.95）
      57.48 （5.51）1.67 （—）1.60 （1.63）1.54 （—）1.78 （1.91）1.37 （1.22）
      610.89 （6.97）1.64 （—）2.34 （1.29）1.66 （—）2.02 （1.41）1.25 （0.74）
      711.03 （7.43）1.69 （—）1.80 （0.90）1.56 （—）2.88 （2.71）1.35 （0.90）
      814.99 （9.01）1.58 （—）3.76 （2.52）2.97 （—）3.96 （2.89）1.33 （1.01）
      97.92 （3.98）1.87 （—）1.62 （1.19）1.74 （—）2.11 （1.13）1.18 （0.65）
      107.30 （6.35）1.97 （—）1.57 （1.54）1.59 （—）2.03 （1.44）1.16 （0.73）

    • Table 2. TRE for different algorithms on COPDgene dataset

      View table

      Table 2. TRE for different algorithms on COPDgene dataset

      CaseInitialVoxelMorphLapIRNGraphregnetOurs
      Mean23.367.984.991.761.53
      126.339.956.851.711.60
      221.799.966.902.752.11
      312.644.411.511.421.21
      429.587.086.382.061.67
      530.089.196.811.811.63
      628.468.124.191.431.35
      721.607.102.731.641.27
      826.467.924.321.541.43
      914.866.933.601.451.38
      1021.819.166.591.791.63

    • Table 3. TRE for different algorithms on Creatis dataset

      View table

      Table 3. TRE for different algorithms on Creatis dataset

      CaseInitialRPMD-FCNGraphregnetOurs
      Mean8.15 （5.60）2.47 （2.27）1.50 （0.85）1.07 （0.73）1.00 （0.69）
      16.34 （2.95）1.84 （1.56）1.34 （0.47）0.87 （0.41）0.86 （0.38）
      214.04 （7.20）3.88 （2.91）1.74 （1.10）1.25 （0.93）1.21 （0.88）
      37.67 （5.05）2.69 （2.66）1.57 （0.87）1.02 （0.60）0.94 （0.73）
      47.33 （4.89）1.89 （1.85）1.64 （0.98）1.11 （0.77）0.95 （0.57）
      57.09 （5.10）2.54 （2.24）1.26 （0.84）1.18 （0.92）1.10 （0.89）
      66.68 （3.68）2.01 （1.49）1.45 （0.81）0.98 （0.51）0.94 （0.46）

    • Table 4. Ablation experiments on DIR-lab, COPDgene, and Creatis datasets

      View table

      Table 4. Ablation experiments on DIR-lab, COPDgene, and Creatis datasets

      MethodTRE /mm
      BaseCorrelation coefficientMulti-scale residual convolutionDense graph convolutionDIR-labCOPDgeneCreatis
      √1.30 （0.86）1.96 （2.41）1.07 （0.73）
      √√1.27 （0.88）1.75 （2.05）1.06 （0.73）
      √√√1.23 （0.84）1.55 （1.70）1.01 （0.69）
      √√√√1.21 （0.83）1.53 （1.42）1.00 （0.69）

    • Table 5. Alignment degree of boundaries and topological preservation capability for different algorithms

      View table

      Table 5. Alignment degree of boundaries and topological preservation capability for different algorithms

      DatasetMethodDSCHD95Jφ<0
      DIR-labVoxelMorph0.945 （0.023）9.126 （3.668）0.11
      LapIRN0.951 （0.019）6.519 （4.289）0.12
      Graphregnet0.954 （0.017）5.868 （6.059）0.12
      Ours0.967 （0.017）4.755 （5.940）0
      COPDgeneVoxelMorph0.917 （0.026）12.861 （3.645）0.20
      LapIRN0.936 （0.028）11.004 （5.157）0.17
      Graphregnet0.942 （0.009）7.503 （4.001）0.15
      Ours0.957 （0.008）6.446 （4.166）0
      CreatisVoxelMorph0.956 （0.028）5.172 （2.814）0.08
      LapIRN0.962 （0.024）2.938 （2.744）0
      Graphregnet0.965 （0.014）3.005 （0.686）0
      Ours0.970 （0.013）2.540 （0.618）0
    Tools

    Get Citation

    Copy Citation Text

    Jianbing Yi, Xi Chen, Feng Cao, Shuxin Yang, Xin Chen. Design of Large Deformable Lung Image Registration Network with Adaptive Window[J]. Acta Optica Sinica, 2024, 44(19): 1915002

    Download Citation

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

    Category: Machine Vision

    Received: Mar. 27, 2024

    Accepted: May. 20, 2024

    Published Online: Oct. 12, 2024

    The Author Email: Yi Jianbing (yijianbing8@jxust.edu.cn), Yang Shuxin (yangshuxin@jxust.edu.cn)

    DOI:10.3788/AOS240778

    CSTR:32393.14.AOS240778

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology