Table 1 Errors calculated for the single vertebrae.

From: 2-step deep learning model for landmarks localization in spine radiographs

Vertebra

Median error x after step I (%)

Median error x after step II (%)

Median error y after step I (%)

Median error y after step II (%)

Number of samples

T9

3.94

2.95

8.89

5.64

15

T10

2.86

2.50

3.47

2.96

66

T11

2.56

2.50

3.09

2.50

139

T12

2.52

2.32

2.53

2.06

188

L1

2.16

1.97

2.11

1.84

194

L2

1.71

1.56

2.12

1.71

195

L3

1.52

1.59

1.79

1.44

195

L4

1.59

1.52

1.68

1.40

195

L5

2.01

1.86

2.21

2.05

195

S1

3.27

2.44

2.51

2.38

195

Weighted mean

2.20

1.98

2.04

1.68

–

  1. The first 2 columns report the median of the absolute error related to the x and y vertebra dimension (Eqs. 1 and 2). The last row sums up the results to have an overall evaluation (Eqs. 5 and 6).
Back to article page