Table 2. Comparison of parameters, floating point operations (FLOPs), category dice scores, and HD95 for different methods on the MSDPT dataset.
| Model | Parameter (M) | Flops (G) | Dice↑ | |
|---|---|---|---|---|
| Pancreas | Tumor | |||
| 3DUnet | 7.91M | 29.745G | 0.6606 ± 0.0199 | 0.2113 ± 0.0914 |
| SegResNet | 1.18M | 15.430G | 0.6832 ± 0.0351 | 0.2633 ± 0.0842 |
| 3DUX-Net | 53.0M | 631.675G | 0.6673 ± 0.0428 | 0.2889 ± 0.9659 |
| AttentionUnet | 23.63M | 245.605G | 0.6698 ± 0.0561 | 0.2961 ± 0.1263 |
| UNETR | 92.78M | 82.550G | 0.6148 ± 0.0496 | 0.1905 ± 0.0664 |
| nnFormer | 149.25M | 213.485G | 0.6664 ± 0.0559 | 0.2668 ± 0.1224 |
| SwinUNETR | 62.17M | 329.331G | 0.6742 ± 0.0359 | 0.2801 ± 0.0911 |
| Ours | 26.97M | 84.837G | 0.7108 ± 0.0320 | 0.3312 ± 0.1111 |
Note:
Bold values indicate the optimal value of each evaluation index.