Author Affiliations
1School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China2College of Medical Instruments, Shanghai University of Medicine and Health Sciences, Shanghai 201318, Chinashow less
Fig. 1. DNet model structure
Fig. 2. Upsampling process. (a) Before sampling; (b) after sampling
Fig. 3. Local self-attention block
Fig. 4. Global self-attention block
Fig. 5. Prior guidance information extraction process
Fig. 6. Comparison of segmentation results
Fig. 7. Feature heatmaps before and after fusion
Fig. 8. Feature heatmaps before and after fusion
Method | mDice | mJC | mASD | mHD |
---|
Average | MYO | RV | LV |
---|
FCN | 87.65 | 86.57 | 83.51 | 92.87 | 0.78 | 0.59 | 3.06 | U-Net | 88.31 | 85.97 | 85.03 | 93.94 | 0.79 | 0.56 | 2.84 | TransUNet | 89.56 | 86.42 | 87.54 | 94.71 | 0.81 | 0.54 | 2.79 | Swin-UNet | 87.56 | 82.97 | 85.69 | 94.03 | 0.79 | 0.42 | 1.92 | DNet | 91.36 | 89.07 | 89.37 | 95.64 | 0.84 | 0.34 | 1.62 |
|
Table 1. Comparison of segmentation effects of different methods on ACDC dataset
Strategy | mDice | mJC | mASD | mHD |
---|
Canny | 86.70 | 0.78 | 3.54 | 25.43 | Sobel | 89.40 | 0.82 | 0.48 | 2.42 | Fourier | 91.36 | 0.84 | 0.34 | 1.62 |
|
Table 2. Comparison of segmentation effects of different prior guided strategies
Strategy | mDice | mJC | mASD | mHD |
---|
U-Net++ | 89.43 | 0.81 | 0.44 | 2.33 | RefineNet | 90.19 | 0.83 | 0.54 | 3.26 | DNet | 91.36 | 0.84 | 0.34 | 1.62 |
|
Table 3. Comparison of segmentation effects of different feature fusion enhancement strategies
Method | mDice | mJC | mASD | mHD |
---|
DNet-PGN-FFEB | 89.79 | 0.82 | 0.62 | 2.51 | DNet-PGN | 90.31 | 0.82 | 0.48 | 2.54 | DNet-FFEB | 90.05 | 0.83 | 0.55 | 2.20 | DNet | 91.36 | 0.84 | 0.34 | 1.62 |
|
Table 4. Results of ablation experiments
Method | mDice | mJC | mASD | mHD |
---|
not concatenated | 87.41 | 0.78 | 0.53 | 2.63 | concatenated | 91.36 | 0.84 | 0.34 | 1.62 |
|
Table 5. Results of prior feature concatenation
Method | mDice | mJC | mASD | mHD |
---|
local attention | 90.16 | 0.83 | 0.43 | 2.17 | global attention | 88.38 | 0.80 | 1.10 | 5.49 | both | 91.36 | 0.84 | 0.34 | 1.62 |
|
Table 6. Results of using local and global attention individually and in combination
Shape | mDice | mJC | mASD | mHD |
---|
round | 91.36 | 0.84 | 0.34 | 1.62 | square | 90.89 | 0.83 | 0.30 | 1.23 |
|
Table 7. Results of using different mask shapes
Radius | mDice | mJC | mASD | mHD |
---|
20 | 90.07 | 0.83 | 0.38 | 2.07 | 30 | 91.36 | 0.84 | 0.34 | 1.62 | 40 | 90.47 | 0.83 | 0.34 | 1.39 |
|
Table 8. Results of different mask radii