Table 3.
Comparison of different deformable registration methods. Standard deviation is provided in the brackets. The best performances are shown in bold
| Methods | ChestX-ray8 | ||||
| MSE | MI | Dice | CPU time/s | GPU time/s | |
| Affine | 0.032 (0.018) | 1.038 (0.188) | 0.922 (0.029) | - | - |
| Affine + SyN | 0.033 (0.014) | 1.317 (0.226) | 0.945 (0.056) | 2.300 (0.042) | - |
| Affine + FFD | 0.023 (0.016) | 1.331 (0.196) | 0.953 (0.030) | 2.358 (0.109) | - |
| Affine + RCN | 0.024 (0.020) | 1.323 (0.129) | 0.951 (0.024) | 0.781 (0.052) | 0.030 (0.002) |
| Affine + RRN | 0.025 (0.017) | 1.313 (0.198) | 0.942 (0.020) | 0.541 (0.163) | 0.022 (0.002) |
| Affine + BUB | 0.021 (0.020) | 1.346 (0.204) | 0.961 (0.023) | 0.284 (0.011) | 0.012 (0.001) |
| Methods | Test12 | ||||
| MSE | MI | Dice | CPU time/s | GPU time/s | |
| Affine | 0.038 (0.043) | 1.235 (0.113) | 0.923 (0.042) | - | - |
| Affine + SyN | 0.034 (0.025) | 1.553 (0.106) | 0.947 (0.025) | 2.245 (0.052) | - |
| Affine + FFD | 0.026 (0.026) | 1.562 (0.192) | 0.954 (0.030) | 2.278 (0.143) | - |
| Affine + RCN | 0.034 (0.046) | 1.560 (0.031) | 0.949 (0.010) | 0.840 (0.206) | 0.031 (0.002) |
| Affine + RRN | 0.035 (0.043) | 1.547 (0.043) | 0.950 (0.032) | 0.457 (0.006) | 0.023 (0.001) |
| Affine + BUB | 0.028 (0.037) | 1.570 (0.138) | 0.960 (0.054) | 0.280 (0.011) | 0.013 (0.001) |