Opto-Electronic Engineering, Volume. 51, Issue 10, 240179(2024)

Colorectal polyp segmentation method combining polarized self-attention and Transformer

Bin Xie1, Yangqian Liu1、*, and Yulin Li2
Author Affiliations
  • 1School of Information Engineering,Jiangxi University of Science and Technology,Ganzhou,Jiangxi 341000,China
  • 2School of Electrical Engineering and Automation,Jiangxi University of Science and Technology,Ganzhou,Jiangxi 341000,China
  • show less
    Figures & Tables(16)
    Colorectal polyp segmentation network combining polarized self-attention and Transformer
    Phase-aware hybrid module
    Segmentation results obtained with/without PAHM
    Polarized self-attention module
    Segmentation results with or without PSA
    Cross-cue fusion module
    Segmentation results obtained with or without CCF
    Visualization of segmentation results of different network models on CVC-ClinicDB and Kvasir datasets
    Visualization of segmentation results of different network models on CVC-ColonDB and ETIS
    • Table 1. Comparison with/without PAHM on CVC-ClinicDB and CVC-ColonDB

      View table
      View in Article

      Table 1. Comparison with/without PAHM on CVC-ClinicDB and CVC-ColonDB

      DatasetMethodDiceMIoUSE
      CVC-ClinicDBN10.9420.8980.950
      N40.9460.9010.951
      CVC-ColonDBN10.8000.7270.819
      N40.8050.7290.822
    • Table 2. Comparison with/without PSA on CVC-ClinicDB and CVC-ColonDB

      View table
      View in Article

      Table 2. Comparison with/without PSA on CVC-ClinicDB and CVC-ColonDB

      DatasetMethodDiceMIoUSE
      CVC-ClinicDBN20.9370.8810.946
      N40.9460.9010.951
      CVC-ColonDBN20.7880.7110.813
      N40.8050.7290.822
    • Table 3. Comparison with/without CCF on CVC-ClinicDB and CVC-ColonDB

      View table
      View in Article

      Table 3. Comparison with/without CCF on CVC-ClinicDB and CVC-ColonDB

      DatasetMethodDiceMIoUSE
      CVC-ClinicDBN30.9420.8940.949
      N40.9460.9010.951
      CVC-ColonDBN30.7510.6840.777
      N40.8050.7290.822
    • Table 4. Experimental parameter settings

      View table
      View in Article

      Table 4. Experimental parameter settings

      DatasetTraindataTestdataPicture size/pixel
      CVC-ClinicDB55062352×352
      Kvasir900100352×352
      ETIS-LaribPolypDB0196352×352
      CVC-ColonDB0380352×352
    • Table 5. Comparison of different algorithms on CVC-ClinicDB and Kvasir

      View table
      View in Article

      Table 5. Comparison of different algorithms on CVC-ClinicDB and Kvasir

      DatasetMethodDiceMIoUSEPCF2MAE
      CVC- ClinicDBU-Net0.8220.7560.8360.8350.8280.020
      PraNet0.9020.8500.9110.9050.9010.009
      EU-Net0.9050.8490.9560.8810.9270.011
      DCRNet0.8990.8470.9120.8930.9070.010
      SSFormer-S0.9190.8720.9030.9390.9080.007
      MSRAFormer0.9340.8840.9500.9240.9440.007
      Ours0.9460.9010.9570.9430.9490.005
      KvasirU-Net0.8210.7470.8550.8560.8280.055
      PraNet0.9010.8410.9100.9160.9030.030
      EU-Net0.9110.8580.9310.9120.9190.028
      DCRNet0.8890.8230.9030.9020.8920.034
      SSFormer-S0.9250.8760.9170.9440.9210.020
      MSRAFormer0.9190.8700.9210.9380.9180.020
      Ours0.9270.8800.9320.9500.9230.020
    • Table 6. Comparison of different algorithms on CVC-ColonDB and ETIS-LaribPolypDB

      View table
      View in Article

      Table 6. Comparison of different algorithms on CVC-ColonDB and ETIS-LaribPolypDB

      DatasetMethodDiceMIoUSEPCF2MAE
      CVC- ColonDBU-Net0.5120.4380.5240.6210.5100.059
      PraNet0.7170.6410.7400.7550.7160.044
      EU-Net0.7560.6830.8480.7560.7890.043
      DCRNet0.7070.6320.7770.7190.7230.051
      SSFormer-S0.7750.6980.7760.8360.7670.034
      MSRAFormer0.7650.6950.8010.8700.7720.031
      Ours0.8050.7290.8780.8720.8060.025
      ETIS- LaribPolypDBU-Net0.4060.3340.4820.4390.4280.037
      PraNet0.6310.5670.6890.6280.6490.030
      EU-Net0.6900.6110.8710.6370.7490.065
      DCRNet0.5480.4840.7440.5040.6000.095
      SSFormer-S0.7700.6950.8560.7440.7820.017
      MSRAFormer0.7490.6740.8210.7870.7820.012
      Ours0.7810.7060.8740.8080.8070.011
    • Table 7. Ablation of each module on Kvasir and EITS datasets

      View table
      View in Article

      Table 7. Ablation of each module on Kvasir and EITS datasets

      MethodCCFPAHMPSAKvasirETIS
      DiceMIoUSEF2DiceMIoUSEF2
      M1×0.9190.8730.9190.9200.7440.6740.8030.769
      M2×0.9180.8720.9130.9140.7400.6720.8020.767
      M3×0.9240.8760.9240.9180.7560.6810.8360.792
      M40.9270.8800.9260.9230.7810.7060.8740.807
    Tools

    Get Citation

    Copy Citation Text

    Bin Xie, Yangqian Liu, Yulin Li. Colorectal polyp segmentation method combining polarized self-attention and Transformer[J]. Opto-Electronic Engineering, 2024, 51(10): 240179

    Download Citation

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

    Category: Article

    Received: Jul. 30, 2024

    Accepted: Sep. 19, 2024

    Published Online: Jan. 2, 2025

    The Author Email: Yangqian Liu (刘阳倩)

    DOI:10.12086/oee.2024.240179

    Topics