Table 5.
Predictive performance of different network architectures, with the mean values listed.
| Methods | 3DIRCADb | SWMH | |||||
|---|---|---|---|---|---|---|---|
| Liver and tumor | Stroke and WMH | ||||||
| DC | VOE | RVD | ASSD | HD | DC | HD | |
| LLRHNet1 | 93.51 & 89.17 | 6.01 & 12.37 | 0.81 & 7.64 | 1.57 & 4.16 | 1.21 & 4.72 | 69.76 & 65.31 | 3.89 & 4.61 |
| LLRHNet2 | 96.36 & 92.31 | 4.97 & 9.26 | 0.80 & 3.08 | 1.05 & 2.76 | 1.34 & 4.37 | 74.83 & 72.19 | 3.87 & 3.08 |
| LLRHNet3 | 96.29 & 91.43 | 5.30 & 8.86 | 0.91 & 4.36 | 0.97 & 2.43 | 2.37 & 3.54 | 75.04 & 73.47 | 3.41 & 2.98 |
| LLRHNet4 | 98.64 & 95.06 | 3.13 & 6.04 | 0.01 & 0.43 | 0.28 & 0.58 | 2.03 & 0.71 | 79.10 & 78.02 | 2.70 & 2.27 |
LLRHNet1 denotes ResNet, LLRHNet2 denotes ResNet only with iterative aggregation, LLRHNet3 denotes ResNet only with transformer, and LLRHNet4 is the final model we choose. The best results are shown in bold. DC:%, HD:mm, VOE:%, RVD:%, HD:mm.