Table 4. Comparison of dice and IOU metrics for different methods on the NIHP dataset.
| Model | Dice↑ | Iou↑ | HD95↓ | ||||
|---|---|---|---|---|---|---|---|
| Background | Pancreas | Mean | Background | Pancreas | Mean | Pancreas | |
| 3DUnet | 0.9995 ± 0.0001 | 0.6739 ± 0.0570 | 0.8367 ± 0.0286 | 0.9990 ± 0.0001 | 0.5220 ± 0.0642 | 0.7605 ± 0.0322 | 37.7600 ± 17.1593 |
| DynUNet | 0.9995 ± 0.0001 | 0.7250 ± 0.0279 | 0.8623 ± 0.0140 | 0.9991 ± 0.0001 | 0.5832 ± 0.0287 | 0.7911 ± 0.0144 | 36.6011 ± 27.1476 |
| AttentionUnet | 0.9996 ± 0.0001 | 0.7109 ± 0.0621 | 0.8552 ± 0.0311 | 0.9991 ± 0.0002 | 0.5681 ± 0.0720 | 0.7836 ± 0.0361 | 21.7137 ± 11.5158 |
| SwinUNETR | 0.9995 ± 0.0001 | 0.7012 ± 0.0272 | 0.8504 ± 0.0136 | 0.9990 ± 0.0001 | 0.5545 ± 0.0316 | 0.7768 ± 0.0159 | 46.2683 ± 19.8101 |
| Ours | 0.9996 ± 0.0001 | 0.7709 ± 0.0264 | 0.8852 ± 0.0132 | 0.9992 ± 0.0001 | 0.6357 ± 0.0306 | 0.8175 ± 0.0153 | 15.3267 ± 7.3831 |
Note:
Bold values indicate the optimal value of each evaluation index.