Table 2.
Impact of kernel size of the global attention unit on model performance
| Organ | Model | Kernel size | DSC (%) |
|---|---|---|---|
| Prostate | ResNet50-UNet-ViT | Conv 3*3 | 88.10 ± 0.63 |
| Conv 5*5 | 89.42 ± 0.89 | ||
| Conv 7*7 | 90.02 ± 1.00 | ||
| VGG16-UNet-ViT | Conv 3*3 | 90.84 ± 1.74 | |
| Conv 5*5 | 91.05 ± 1.22 | ||
| Conv 7*7 | 91.75 ± 1.36 | ||
| Bladder | ResNet50-UNet-ViT | Conv 3*3 | 94.22 ± 0.63 |
| Conv 5*5 | 95.04 ± 1.21 | ||
| Conv 7*7 | 94.98 ± 0.83 | ||
| VGG16-UNet-ViT | Conv 3*3 | 91.46 ± 1.36 | |
| Conv 5*5 | 91.36 ± 1.00 | ||
| Conv 7*7 | 95.32 ± 0.96 | ||
| Rectum | ResNet50-UNet-ViT | Conv 3*3 | 83.84 ± 1.21 |
| Conv 5*5 | 84.26 ± 0.72 | ||
| Conv 7*7 | 83.86 ± 1.69 | ||
| VGG16-UNet-ViT | Conv 3*3 | 84.11 ± 0.81 | |
| Conv 5*5 | 86.28 ± 1.21 | ||
| Conv 7*7 | 87.00 ± 1.97 | ||
| RFH | ResNet50-UNet-ViT | Conv 3*3 | 95.49 ± 0.85 |
| Conv 5*5 | 94.87 ± 1.00 | ||
| Conv 7*7 | 95.83 ± 0.95 | ||
| VGG16-UNet-ViT | Conv 3*3 | 94.11 ± 0.65 | |
| Conv 5*5 | 96.04 ± 1.24 | ||
| Conv 7*7 | 96.30 ± 0.65 | ||
| LFH | ResNet50-UNet-ViT | Conv 3*3 | 95.04 ± 0.81 |
| Conv 5*5 | 94.29 ± 0.77 | ||
| Conv 7*7 | 94.80 ± 0.97 | ||
| VGG16-UNet-ViT | Conv 3*3 | 95.07 ± 0.59 | |
| Conv 5*5 | 95.27 ± 1.16 | ||
| Conv 7*7 | 96.34 ± 0.63 |
The best performance for each organ is highlighted in bold