Table 5 Comparison of the performance of various CNN backbones on ISIC 2019 and HAM10000 datasets.

From: A deep learning framework for automated early diagnosis and classification of skin cancer lesions in dermoscopy images

Network Backbones

ISIC 2019

HAM1000

AR

Pr.

Re.

F1

AUC

AR

Pr.

Re.

F1

AUC

ResNet34

75.67

75.78

74.87

75.32

79.99

80.41

78.89

77.97

78.45

75.59

ResNet50

87.56

87.67

88.23

87.94

86.23

88.96

88.87

88.73

88.80

88.33

ResNet101

95.94

95.23

96.94

96.07

96.78

97.14

97.12

96.84

96.98

97.18

DenseNet121

91.34

93.34

95.45

94.38

91.56

94.55

94.45

95.19

94.82

94.28

InceptionV3

88.89

88.34

87.23

87.78

85.23

90.67

91.61

91.78

91.69

91.34

HRNet

97.54

97.73

98.14

97.93

97.28

98.34

98.56

98.51

98.53

98.23

EfficientNetB0

96.74

98.13

96.64

97.38

96.18

97.67

98.89

96.98

97.93

97.77

ConvNeXtBase

95.14

95.13

96.14

95.63

96.88

97.78

97.92

97.23

97.57

97.45

Back to article page