Skip to main content

Table 2 The performance of DCNNs that including InceptionV3, InceptionResnetV2, DenseNet169, and ResNet50 in the internal and external datasets

From: Automatic differentiation of thyroid scintigram by deep convolutional neural network: a dual center study

Metrics

Dataset

Diffusely increased (%)

Diffusely decreased (%)

Focal increased (%)

Heterogeneous uptake (%)

InceptionV3

Accuracy

Internal

94.51

99.68

98.38

92.89

External

89.74

99.34

98.68

87.75

Sensitivity (Recall)

Internal

90.77

99.56

100.00

81.48

External

90.00

95.92

76.92

83.33

Specificity

Internal

97.21

99.75

98.32

95.30

External

89.56

100.00

99.65

90.66

PPV(Precision)

Internal

95.93

99.56

71.43

78.57

External

85.04

100.00

90.91

85.47

NPV

Internal

93.57

99.75

100.00

96.06

External

93.14

99.22

98.97

89.19

F1 score

Internal

93.28

99.56

83.33

80.00

External

87.45

97.92

83.33

84.39

InceptionResnetV2

Accuracy

Internal

94.02

99.35

98.06

92.41

External

90.40

97.02

99.01

87.09

Sensitivity (Recall)

Internal

91.15

99.12

88.00

79.63

External

90.83

81.63

84.62

85.00

Specificity

Internal

96.10

99.49

98.48

95.11

External

90.11

100.00

99.65

88.46

PPV(Precision)

Internal

94.42

99.12

70.97

77.48

External

85.83

100.00

91.67

82.93

NPV

Internal

93.75

99.49

99.49

95.67

External

93.71

96.56

99.31

89.94

F1 score

Internal

92.76

99.12

78.57

78.54

External

88.26

89.89

88.00

83.95

DenseNet169

Accuracy

Internal

94.83

99.03

95.32

91.44

External

91.06

98.68

99.01

91.39

Sensitivity (Recall)

Internal

92.69

97.79

100.00

66.67

External

95.00

91.84

100.00

83.33

Specificity

Internal

96.38

99.75

95.12

96.67

External

88.46

100.00

98.96

96.70

PPV(Precision)

Internal

94.88

99.55

46.30

80.90

External

84.44

100.00

81.25

94.34

NPV

Internal

94.79

98.74

100.00

93.21

External

96.41

98.44

100.00

89.80

F1 score

Internal

93.77

98.66

63.29

73.10

External

89.41

95.74

89.66

88.50

ResNet50

Accuracy

Internal

93.70

99.52

98.55

92.08

External

91.39

99.01

98.68

89.74

Sensitivity (Recall)

Internal

93.46

99.56

96.00

71.30

External

95.00

93.88

76.92

83.33

Specificity

Internal

93.87

99.49

98.65

96.48

External

89.01

100.00

99.65

93.96

PPV(Precision)

Internal

91.70

99.12

75.00

81.05

External

85.07

100.00

90.91

90.09

NPV

Internal

95.20

99.74

99.83

94.08

External

96.43

98.83

98.97

89.53

F1 score

Internal

92.57

99.34

84.21

75.86

External

89.76

96.84

83.33

86.58