Table 3 Root-mean-square percentage coefficient of variation (RMS %CV) and least significant change (LSC) compared between the proposed algorithm (U-Net) and the current semi-automated gold standard (Standard) segmentation protocols on the held-out precision dataset.

From: Automatic segmentation of trabecular and cortical compartments in HR-pQCT images using an embedding-predicting U-Net and morphological post-processing

  

RMS %CV

LSC

Wilcoxon

p Value

U-Neta

Standardb

U-Net

Standard

Radius (n = 71)

Tt.BMD

mg HA/cm3

0.29

0.70

2.51

5.60

0.796

Ct.BMD

mg HA/cm3

0.27

0.79

6.67

19.12

0.010c

Tb.BMD

mg HA/cm3

0.51

0.61

1.87

2.37

0.055

Ct.Th

mm

0.69

0.80

0.022

0.025

0.781

Ct.Po

%

7.92

9.55

0.23

0.25

0.001c

Tb.BV/TV

%

1.00

1.06

0.51

0.54

0.030c

Tb.N

mm−1

1.40

1.41

0.057

0.057

0.030c

Tb.Th

mm

0.63

0.72

0.004

0.005

0.005c

Tb.Sp

mm

1.06

1.07

0.021

0.022

0.036c

Tt.Ar

mm2

0.71

0.77

6.20

5.01

0.000c

Ct.Ar

mm2

0.76

1.17

1.46

1.78

0.000c

Tb.Ar

mm2

0.79

1.27

5.35

6.29

0.284

Tibia (n = 85)

Tt.BMD

mg HA/cm3

0.47

0.48

3.72

3.74

0.639

Ct.BMD

mg HA/cm3

0.36

0.45

8.37

10.47

0.000c

Tb.BMD

mg HA/cm3

0.61

0.69

2.79

3.06

0.016c

Ct.Th

mm

0.61

0.74

0.026

0.032

0.333

Ct.Po

%

9.07

11.67

0.76

0.82

0.045c

Tb.BV/TV

%

0.83

0.82

0.50

0.52

0.560

Tb.N

mm−1

2.97

2.98

0.11

0.11

0.246

Tb.Th

mm

0.56

0.67

0.004

0.005

0.460

Tb.Sp

mm

2.21

2.21

0.045

0.045

0.497

Tt.Ar

mm2

0.27

0.13

5.63

2.74

0.000d

Ct.Ar

mm2

0.37

0.66

1.41

2.42

0.991

Tb.Ar

mm2

0.29

0.16

4.92

2.68

0.000d

  1. aU-Net: automated segmentation algorithm using a U-Net and morphological post-processing.
  2. bStandard: Current standard semi-automated segmentation protocol.
  3. cWilcoxon signed-rank test indicates significantly lower individual standard deviations with automated segmentation algorithm as compared to the standard semi-automated protocol.
  4. dWilcoxon signed-rank test indicates significantly higher individual standard deviations with automated segmentation algorithm as compared to the standard semi-automated protocol.
Back to article page