Table 2.
Results for our experiments comparing single-level and two-level warping approach as well as cross-validation and training without withheld data: Dice scores listed by anatomical structures of our experiments using Cycle Discrepancy for training.
Liver | Spleen | Lkidney | Rkidney | Mean | |
---|---|---|---|---|---|
initial | 59.32 | 36.90 | 36.59 | 37.02 | 42.46 |
± 14.03 | ± 19.49 | ± 19.53 | ± 22.08 | ± 18.78 | |
cross-validation | 77.95 | 69.89 | 70.18 | 71.85 | 72.30 |
1 warp | ± 8.16 | ± 16.00 | ± 24.34 | ± 34.40 | ± 20.75 |
cross-validation | 80.71 | 72.12 | 79.33 | 74.65 | 76.68 |
2 warps | ± 9.33 | ± 17.08 | ± 26.06 | ± 36.91 | ± 22.34 |
trained without withheld data | 81.04 | 71.11 | 76.27 | 76.49 | 76.23 |
1 warp | ± 8.22 | ± 18.03 | ± 24.25 | ± 32.64 | ±20.88 |
trained without withheld data | 81.85 | 76.77 | 79.81 | 80.17 | 79.65 |
2 warps | ± 0.58 | ± 13.64 | ± 24.52 | ± 34.65 | ± 20.25 |
NiftyReg | 83.97 | 76.55 | 79.83 | 79.26 | 79.90 |
reg_aladin | ± 6.19 | ± 12.00 | ± 7.12 | ± 37.55 | ± 15.15 |