Table 1.
Comparison with state-of-the-art methods (mean ± SD).
| Model | ET dice (%) | TC dice (%) | WT dice (%) |
|---|---|---|---|
| 3D U-Net3 | 82.3 ± 3.8 | 85.6 ± 3.2 | 89.4 ± 2.6 |
| Attention U-Net10 | 83.8 ± 3.5 | 86.9 ± 3.0 | 90.5 ± 2.4 |
| nnU-Net23 | 85.7 ± 3.1 | 88.4 ± 2.6 | 91.6 ± 2.1 |
| ConvLSTM-Longitudinal5,24 | 84.9 ± 3.4 | 87.5 ± 2.8 | 91.0 ± 2.3 |
| UNETR3 | 85.4 ± 3.0 | 88.1 ± 2.7 | 91.8 ± 2.0 |
| Swin-UNet3 | 86.2 ± 2.9 | 88.9 ± 2.5 | 92.1 ± 1.9 |
| Static–Dynamic Transformer6 | 86.5 ± 2.8 | 89.0 ± 2.4 | 92.3 ± 1.8 |
| Hybrid CNN–Transformer25 | 86.9 ± 2.6 | 89.3 ± 2.3 | 92.4 ± 1.7 |
| PGSMT (proposed) | 88.1 ± 2.1 | 90.2 ± 1.9 | 93.0 ± 1.4 |