Table 1 Comparison between our method and baselines.

From: Neural architecture search for pneumonia diagnosis from chest X-rays

Model

Sensitivity (%)

Specificity (%)

F1 (%)

AUC (%)

Accuracy (%)

Model size

Training time (h)

Inference time (ms)

VGG1951

92.7 ± 0.68

92.4 ± 0.93

93.0 ± 0.59

93.9 ± 0.81

92.7 ± 0.84

731

2.3

69.2

InceptionV352

91.8 ± 0.49

92.2 ± 0.70

91.4 ± 0.76

92.8 ± 0.55

92.6 ± 0.92

502

2.1

38.6

DenseNet12152

93.8 ± 0.87

91.7 ± 0.92

92.4 ± 0.96

93.8 ± 0.53

93.1 ± 0.97

537

2.1

87.2

AlexNet52

92.5 ± 1.04

92.7 ± 0.85

92.1 ± 0.62

94.1 ± 0.62

92.7 ± 0.73

433

2.0

32.7

VGG1651

90.9 ± 0.75

94.1 ± 0.68

91.8 ± 1.15

94.3 ± 0.47

92.5 ± 0.61

737

2.2

55.3

Xception51

90.7 ± 0.59

92.3 ± 0.71

93.6 ± 0.74

93.4 ± 0.62

92.1 ± 0.73

241

1.8

146.9

GoogLeNet52

90.7 ± 1.03

92.5 ± 0.91

91.8 ± 0.72

95.4 ± 0.37

93.4 ± 0.85

87

1.5

38.1

LeNet553

84.6 ± 0.72

85.9 ± 0.55

85.4 ± 0.59

88.7 ± 0.36

89.1 ± 0.42

11.1

0.2

28.0

Kermany et al.10

92.8 ± 0.59

92.2 ± 0.57

92.5 ± 0.96

93.7 ± 0.69

93.0 ± 0.68

403

2.2

172.6

Stephen et al.30

92.4 ± 0.71

92.7 ± 0.38

92.4 ± 0.96

94.2 ± 0.71

93.7 ± 0.62

61

1.6

147.0

Siddiqi31

94.7 ± 0.42

93.1 ± 1.33

92.7 ± 0.61

93.9 ± 0.33

93.5 ± 0.74

274

1.7

210.6

Liang et al.27

89.5 ± 0.62

91.7 ± 0.73

89.9 ± 1.04

92.2 ± 0.68

92.3 ± 0.95

\(\approx \) 215

1.9

187.3

Meta Pseudo Label39

90.6 ± 0.74

92.3 ± 0.58

91.7 ± 0.94

93.2 ± 0.63

91.8 ± 0.71

69

1.7

162.5

Liu et al.40

92.0 ± 0.58

92.7 ± 0.84

92.4 ± 0.81

93.4 ± 0.45

92.4 ± 0.74

35

1.4

85.2

Kundu et al.54

92.4 ± 1.05

91.6 ± 0.69

91.8 ± 0.95

93.1 ± 0.52

91.9 ± 0.50

195

2.1

141.7

Cha et al.55

92.1 ± 0.62

91.3 ± 0.62

91.4 ± 0.77

93.2 ± 0.68

92.0 ± 0.59

131

1.7

196.3

DARTS7

88.9 ± 0.71

89.2 ± 0.95

90.1 ± 0.62

93.0 ± 0.85

89.8 ± 0.75

11.4

0.9

28.7

LBT-DARTS (ours)

93.0 ± 0.42

93.2 ± 0.86

92.8 ± 0.75

94.9 ± 0.82

93.3 ± 0.61

11.2

0.9

28.5

PC-DARTS8

93.2 ± 0.84

90.9 ± 0.95

91.8 ± 0.62

92.5 ± 0.60

91.4 ± 0.75

11.3

0.1

28.5

LBT-PC-DARTS (ours)

95.9 ± 0.74

96.7 ± 0.92

97.1 ± 0.64

97.6 ± 0.58

97.0 ± 0.80

10.9

0.1

26.4

  1. Model size is in MB. Training time is in GPU hours (h). Inference time is in milliseconds (ms).
  2. Significant values are in bold.