Table 4 Comparisons with the SOTA methods on 3DIRCADb test datasets.

From: Deep supervision and atrous inception-based U-Net combining CRF for automatic liver segmentation from CT

Model

Method

Dice (%)

VOE (%)

RVD (%)

ASD (mm)

RMSD (mm)

Christ et al.3

2D Cascaded FCN

90.23 ± 2.65

14.28 ± 4.58

− 2.55 ± 1.22

7.21 ± 3.95

10.22 ± 3.96

Chlebus et al.30

2D U-Net

93.36 ± 1.63

10.32 ± 3.12

− 1.19 ± 1.01

5.32 ± 3.01

7.45 ± 6.25

Han et al.6

2D ResNet

93.85 ± 1.25

9.55 ± 2.11

− 1.02 ± 0.98

5.98 ± 3.12

7.22 ± 7.12

Seo et al.13

2D mU-Net

96.25 ± 1.01

8.45 ± 2.02

0.97 ± 0.24

3.87 ± 1.21

6.25 ± 2.01

Li et al.11

3D H-DenseUNet

98.74 ± 0.21

7.47 ± 2.12

0.16 ± 0.09

1.22 ± 1.11

2.85 ± 3.11

Proposed

3D DA-UNet + CRF

98.17 ± 0.19

3.58 ± 0.38

0.18 ± 0.12

0.95 ± 1.31

2.57 ± 0.32

  1. Significant values are in bold.
Back to article page