Acta Optica Sinica, Volume. 45, Issue 7, 0717001(2025)

LTDA‐Mamba: Retinal Vessel Segmentation Based on a Hybrid CNN‐Mamba Network

Yuanyuan Peng1,*... Haoyang Li1, Wen Li1 and Yuejin Zhang2 |Show fewer author(s)
Author Affiliations
  • 1School of Electrical and Automation Engineering, East China Jiaotong University, Nanchang 330000, Jiangxi , China
  • 2School of Information and Software Engineering, East China Jiaotong University, Nanchang 330000, Jiangxi , China
  • show less
    Figures & Tables(8)
    Framework of proposed LTDA-Mamba
    Specific structures of VSS module, Mamba layer, and ResMamba block[15]. (a) VSS module; (b) Mamba layer; (c) ResMamba block
    Specific architecture of LIOT[10]
    DCA module[14]. (a) DCA block; (b) CCA module; (c) SCA module
    Experimental results with different state-of-the-art models on DRIVE, CHASE_DB1 and STARE datasets. (a) Images; (b) labels; (c) UNet++; (d) CS2Net; (e) SA-UNet; (f) LIOT; (g) LightM-UNet; (h) LTDA-Mamba
    • Table 1. Quantitative evaluation with different methods

      View table

      Table 1. Quantitative evaluation with different methods

      DatasetYearMethodF1 scoreAccuracySensitivitySpecificity
      DRIVE2020UNet++0.79670.96720.72960.9900
      2021CS2Net0.80860.96760.78080.9855
      2021SA-UNet0.80800.96760.77800.9858
      2022LIOT0.79150.96660.72740.9898
      2024LightM-UNet0.79530.96720.72810.9896
      2024Ours0.81510.96890.78680.9867
      CHASE_DB12020UNet++0.68510.96490.55680.9949
      2021CS2Net0.80350.97360.78620.9874
      2021SA-UNet0.79070.97290.74790.9894
      2022LIOT0.80020.97250.81930.9832
      2024LightM-UNet0.80050.97110.78430.9852
      2024Ours0.80430.97410.76970.9894
      STARE2020UNet++0.65940.95970.50820.9972
      2021CS2Net0.79100.96970.74720.9882
      2021SA-UNet0.77430.96770.72280.9880
      2022LIOT0.75410.96770.64530.9945
      2024LightM-UNet0.80490.97340.76990.9911
      2024Ours0.82190.97920.74880.9929
    • Table 2. FLOPs and Params for different methods

      View table

      Table 2. FLOPs and Params for different methods

      MethodUNet++CS2NetSA-UNetLIOTLightM-UNetOurs
      FLOPs /G9051851362505125132
      Params /M36.638.440.4813.600.430.47
    • Table 3. Influence of each module on presented framework

      View table

      Table 3. Influence of each module on presented framework

      DatasetBaselineLIOTDCAF1 scoreAccuracySensitivitySpecificity
      DRIVE××0.79530.96720.72810.9896
      ×0.80920.96670.78520.9866
      ×0.81210.96750.80020.9839
      0.81510.96890.78680.9867
      CHASE_DB1××0.80050.97110.78430.9852
      ×0.80110.97320.80620.9849
      ×0.80240.97370.77310.9894
      0.80430.97410.76970.9894
      STARE××0.80490.97340.76990.9911
      ×0.81610.97460.77110.9906
      ×0.81520.97400.77050.9920
      0.82190.97920.74880.9929
    Tools

    Get Citation

    Copy Citation Text

    Yuanyuan Peng, Haoyang Li, Wen Li, Yuejin Zhang. LTDA‐Mamba: Retinal Vessel Segmentation Based on a Hybrid CNN‐Mamba Network[J]. Acta Optica Sinica, 2025, 45(7): 0717001

    Download Citation

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

    Category: Medical optics and biotechnology

    Received: Dec. 13, 2024

    Accepted: Jan. 16, 2025

    Published Online: Mar. 19, 2025

    The Author Email: Peng Yuanyuan (pengmi467347713@126.com)

    DOI:10.3788/AOS241887

    Topics