Table 8 Results of ablation experiments on key components(%)
From: GH-UNet: group-wise hybrid convolution-VIT for robust medical image segmentation
Datasets | PW | MSGA | GDG | Cascade | Dice↑ | IoU↑ |
|---|---|---|---|---|---|---|
ISIC2016 | 85.32 | 83.81 | ||||
✓ | 91.76 | 86.43 | ||||
✓ | ✓ | 93.32 | 87.61 | |||
✓ | ✓ | ✓ | 93.41 | 87.75 | ||
✓ | ✓ | ✓ | ✓ | 93.56 | 88.01 | |
Kvasir-SEG | 87.82 | 80.85 | ||||
✓ | 90.01 | 83.87 | ||||
✓ | ✓ | 91.91 | 85.87 | |||
✓ | ✓ | ✓ | 92.21 | 86.34 | ||
✓ | ✓ | ✓ | ✓ | 92.46 | 86.81 | |
Synapse | 60.78 | - | ||||
✓ | 69.22 | - | ||||
✓ | ✓ | 75.32 | - | |||
✓ | ✓ | ✓ | 76.87 | - | ||
✓ | ✓ | ✓ | ✓ | 77.46 | - |
