Table 3 Confusion metrics of the classification and regression network for classifying centrilobular and paraseptal emphysema severity scores against the visual scores on the evaluation set (\(n=7143\)).

From: Emphysema subtyping on thoracic computed tomography scans using deep neural networks

(a) Classification network in predicting centrilobular emphysema severity

Predict

Visual scores

Precision (%)

Absent

Trace

Mild

Moderate

Confluence

Advanced

Absent

1621

530

259

21

0

0

66.68

Trace

625

451

322

50

0

0

31.15

Mild

230

294

664

288

4

0

44.86

Moderate

22

42

142

514

141

7

59.22

Confluence

0

3

15

153

322

42

60.19

Advanced

1

2

7

23

189

159

41.73

Recall (%)

64.87

34.11

47.13

49.00

49.09

76.44

52.23 (ACC%)

(b) Classification network in predicting paraseptal emphysema severity

 

Predict

Visual scores

Precision (%)

Absent

Mild

Substantial

Absent

2508

740

181

73.14

Mild

1144

942

467

36.90

Substantial

205

183

773

66.58

Recall (%)

65.02

50.51

54.40

59.12 (ACC%)

(c) Regression network in predicting centrilobular emphysema severity

Predict

Visual scores

Precision (%)

Absent

Trace

Mild

Moderate

Confluence

Advanced

Absent

1734

536

206

17

0

0

69.55

Trace

696

635

682

142

2

0

29.43

Mild

56

121

393

319

21

0

43.19

Moderate

11

26

117

453

192

8

56.13

Confluence

2

3

10

112

334

83

61.09

Advanced

0

1

1

5

107

117

50.65

Recall (%)

69.39

48.03

27.89

43.18

50.91

56.25

51.32 (ACC%)

(d) Regression network in predicting paraseptal emphysema severity

Predict

Visual scores

Precision (%)

Absent

Mild

Substantial

Absent

2976

839

104

75.94

Mild

784

778

455

38.57

Substantial

97

248

862

71.42

Recall (%)

77.16

41.72

60.66

64.62 (ACC%)

  1. Advanced destructive emphysema is denoted as Advanced.
Back to article page