• Laser & Optoelectronics Progress
  • Vol. 62, Issue 2, 0217003 (2025)
Linfeng Kong1,2,* and Yun Wu1,2
Author Affiliations
  • 1State Key Laboratory of Public Big Data, Guizhou University, Guiyang 550025, Guizhou , China
  • 2College of Computer Science and Technology, Guizhou University, Guiyang 550025, Guizhou , China
  • show less
    DOI: 10.3788/LOP241105 Cite this Article Set citation alerts
    Linfeng Kong, Yun Wu. Retinal Vessel Segmentation Using Multi-Directional Stripe Convolution and Pyramid Dual Pooling[J]. Laser & Optoelectronics Progress, 2025, 62(2): 0217003 Copy Citation Text show less
    Overall structure of MPU-Net model
    Fig. 1. Overall structure of MPU-Net model
    Structure of FDSC module in encoder
    Fig. 2. Structure of FDSC module in encoder
    Structure of PDPF2 module
    Fig. 3. Structure of PDPF2 module
    Structure of CSDA module
    Fig. 4. Structure of CSDA module
    Structure of ECA module
    Fig. 5. Structure of ECA module
    Structure of SA module
    Fig. 6. Structure of SA module
    Partial images after preprocessing
    Fig. 7. Partial images after preprocessing
    Visual comparison of segmentation results by different methods
    Fig. 8. Visual comparison of segmentation results by different methods
    Visual comparison of ablation experiments. (a1) (a2) Images from CHASE-DB1 test set; (b1) (b2) images from DRIVE test set
    Fig. 9. Visual comparison of ablation experiments. (a1) (a2) Images from CHASE-DB1 test set; (b1) (b2) images from DRIVE test set
    MethodYearCHASE-DB1 datasetDRIVE dsataset
    AUCF1AccAUCF1Acc
    Ref. [2320180.98390.80310.96560.97930.82020.9561
    Ref. [2420180.98250.96370.98070.9567
    Ref. [1220190.96660.80060.97030.97140.80910.9630
    Ref. [2520190.98120.80370.96610.97720.82700.9567
    Ref. [1420200.97940.81940.9561
    Ref. [2620200.98320.96940.98400.9610
    Ref. [2720210.78930.97110.81310.9660
    Ref. [2820220.75200.96600.80800.9660
    Ref. [2920220.98180.77410.97140.98280.81490.9678
    Ref. [3020220.98580.97320.98320.9675
    Ref. [3120230.90800.78960.97080.89750.81010.9650
    Ref. [3220230.80310.96120.81290.9589
    Ref. [1520230.79920.97370.81550.9668
    MPU-Net20240.98630.80460.97460.98510.81620.9689
    Table 1. Comparison of results by different segmentation methods
    ModelAUCF1Acc
    UNet80.97390.80790.9673
    UNet++90.97780.81360.9682
    ResUnet330.97800.81180.9680
    ACC-UNet340.97970.81110.9681
    MPU-Net (proposed)0.98510.81620.9689
    Table 2. Performance comparison between MPU-Net and other models
    ModelAUCF1Acc
    U-Net0.98130.78700.9718
    U-Net+FDSC0.98420.80390.9735
    U-Net+PDPF20.98090.78830.9730
    U-Net+CSDA0.98070.78870.9723
    MPU-Net0.98630.80460.9746
    Table 3. Results of ablation experiments on CHASE-DB1 dataset
    ModelAUCF1Acc
    U-Net0.97390.80790.9673
    U-Net+FDSC0.98240.81600.9681
    U-Net+PDPF20.97420.80890.9673
    U-Net+CSDA0.97470.80870.9676
    MPU-Net0.98510.81620.9689
    Table 4. Results of ablation experiments on DRIVE dataset