• Laser & Optoelectronics Progress
  • Vol. 61, Issue 10, 1037005 (2024)
Keyan Chen1, Qiaohong Liu2、*, Xiaoxiang Han1, Yuanjie Lin1, and Weikun Zhang1
Author Affiliations
  • 1School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • 2College of Medical Instruments, Shanghai University of Medicine and Health Sciences, Shanghai 201318, China
  • show less
    DOI: 10.3788/LOP231800 Cite this Article Set citation alerts
    Keyan Chen, Qiaohong Liu, Xiaoxiang Han, Yuanjie Lin, Weikun Zhang. Cardiac Image Segmentation by Combining Frequency Domain Prior and Feature Enhancement[J]. Laser & Optoelectronics Progress, 2024, 61(10): 1037005 Copy Citation Text show less
    DNet model structure
    Fig. 1. DNet model structure
    Upsampling process. (a) Before sampling; (b) after sampling
    Fig. 2. Upsampling process. (a) Before sampling; (b) after sampling
    Local self-attention block
    Fig. 3. Local self-attention block
    Global self-attention block
    Fig. 4. Global self-attention block
    Prior guidance information extraction process
    Fig. 5. Prior guidance information extraction process
    Comparison of segmentation results
    Fig. 6. Comparison of segmentation results
    Feature heatmaps before and after fusion
    Fig. 7. Feature heatmaps before and after fusion
    Feature heatmaps before and after fusion
    Fig. 8. Feature heatmaps before and after fusion
    MethodmDicemJCmASDmHD
    AverageMYORVLV
    FCN87.6586.5783.5192.870.780.593.06
    U-Net88.3185.9785.0393.940.790.562.84
    TransUNet89.5686.4287.5494.710.810.542.79
    Swin-UNet87.5682.9785.6994.030.790.421.92
    DNet91.3689.0789.3795.640.840.341.62
    Table 1. Comparison of segmentation effects of different methods on ACDC dataset
    StrategymDicemJCmASDmHD
    Canny86.700.783.5425.43
    Sobel89.400.820.482.42
    Fourier91.360.840.341.62
    Table 2. Comparison of segmentation effects of different prior guided strategies
    StrategymDicemJCmASDmHD
    U-Net++89.430.810.442.33
    RefineNet90.190.830.543.26
    DNet91.360.840.341.62
    Table 3. Comparison of segmentation effects of different feature fusion enhancement strategies
    MethodmDicemJCmASDmHD
    DNet-PGN-FFEB89.790.820.622.51
    DNet-PGN90.310.820.482.54
    DNet-FFEB90.050.830.552.20
    DNet91.360.840.341.62
    Table 4. Results of ablation experiments
    MethodmDicemJCmASDmHD
    not concatenated87.410.780.532.63
    concatenated91.360.840.341.62
    Table 5. Results of prior feature concatenation
    MethodmDicemJCmASDmHD
    local attention90.160.830.432.17
    global attention88.380.801.105.49
    both91.360.840.341.62
    Table 6. Results of using local and global attention individually and in combination
    ShapemDicemJCmASDmHD
    round91.360.840.341.62
    square90.890.830.301.23
    Table 7. Results of using different mask shapes
    RadiusmDicemJCmASDmHD
    2090.070.830.382.07
    3091.360.840.341.62
    4090.470.830.341.39
    Table 8. Results of different mask radii
    Keyan Chen, Qiaohong Liu, Xiaoxiang Han, Yuanjie Lin, Weikun Zhang. Cardiac Image Segmentation by Combining Frequency Domain Prior and Feature Enhancement[J]. Laser & Optoelectronics Progress, 2024, 61(10): 1037005
    Download Citation