Table 2 Quantitative results of segmentation experiments on vertebral body structures for 6 segmentation networks

From: WGAN-based multi-structure segmentation of vertebral cross-section MRI using ResU-Net and clustered transformer

Method

Dice (%)

HD (mm)

mIOU (%)

PPV (%)

U-Net

81.4 ± 1.8

5.2 ± 1.7

88.2 ± 1.3

90.8 ± 1.8

nnU-Net

90.4 ± 1.6

2.5 ± 2.4

94.5 ± 2.5

96.7 ± 2.3

SegNet

81.3 ± 2.5

4.8 ± 1.2

90.6 ± 1.1

94.3 ± 3.4

DeepLab v3+

85.3 ± 1.9

4.2 ± 3.7

85.8 ± 3.3

89.5 ± 1.6

Spine-GAN

89.5 ± 5.1

2.3 ± 1.9

92.1 ± 2.9

96.8 ± 1.3

Our Model

92.6 ± 2.8

1.7 ± 2.4

96.2 ± 1.5

96.8 ± 0.4

  1. (values represent mean ± standard deviation)
Back to article page