Table 3 Generalization performance on the public LiTS and MSD test sets for the segmentation task
LiTS Dataset | MSD Dataset | |||
|---|---|---|---|---|
Model | DSC (%) ↑ | HD95 (mm) ↓ | DSC (%) ↑ | HD95 (mm) ↓ |
CNN-based Methods | ||||
3D U-Net | 84.1 | 20.5 | 85.3 | 18.9 |
V-Net | 84.5 | 19.8 | 85.9 | 18.1 |
nnU-Net | 89.2 | 10.1 | 90.4 | 9.85 |
Transformer-based Methods | ||||
UNETR | 88.5 | 12.3 | 89.6 | 11.5 |
Swin UNETR | 89.6 | 9.88 | 90.9 | 9.50 |
MedNeXt | 90.1 | 9.21 | 91.3 | 8.99 |
SegMamba | 89.8 | 9.65 | 91.0 | 9.31 |
Foundation Model-based Methods | ||||
Medical SAM | 86.2 | 15.4 | 87.5 | 14.8 |
SAM-Med3D | 87.1 | 13.9 | 88.3 | 13.1 |
Ours | ||||
STD-Net | 90.8 | 8.55 | 91.9 | 8.13 |
