Table 3 Registration performance comparison with SOTAs using a one-step learning strategy (without one-cycle transfer training applied) and using a two-step learning strategy (with one-cycle transfer learning applied).

From: A two-step deep learning method for 3DCT-2DUS kidney registration during breathing

 

Metric (mm)

One-step learning

Two-step learning

CT–CT

CT–US

CT–CT

CT–US

Ours

HD

11.59 ± 1.57

6.71 ± 1.06

3.97 ± 1.37

3.80 ± 0.87

MCD

4.09 ± 0.65

2.03 ± 0.29

1.15 ± 0.57

0.94 ± 0.18

VoxelMorph

HD

12.21 ± 1.87

5.99 ± 0.93

8.15 ± 3.94

4.16 ± 0.97

MCD

4.38 ± 0.72

1.82 ± 0.28

2.71 ± 1.48

1.05 ± 0.21

C2FViT

HD

10.36 ± 1.77

6.42 ± 0.84

9.93 ± 1.72

5.64 ± 0.90

MCD

3.74 ± 0.91

1.93 ± 0.50

3.48 ± 0.78

1.45 ± 0.42

VTN-Affine

HD

12.18 ± 1.86

5.58 ± 1.01

7.66 ± 2.95

4.14 ± 1.15

MCD

4.21 ± 0.71

1.45 ± 0.35

2.45 ± 1.10

1.02 ± 0.20

ConvNet-Affine

HD

11.58 ± 2.43

6.72 ± 1.34

4.56 ± 1.55

3.89 ± 0.91

MCD

4.36 ± 1.11

2.07 ± 0.51

1.46 ± 0.62

1.00 ± 0.23

  1. Significant values are in bold.