Table 4 Diagnostic performance of 157 dermatologists by MClass-D dataset.

From: AI outperformed every dermatologist in dermoscopic melanoma diagnosis, using an optimized deep-CNN architecture with custom mini-batch logic and loss function

Subset of dermatologists

AUC

SEN

SPE

All participants (n = 157)

67.1

74.1

60.0

University hospital (n \(=\) 151)

66.9

74.0

59.8

Private practice (resident) (n \(=\) 6)

71.3

76.7

65.8

Position in hospital hierarchy

Junior physicians (n \(=\) 88)

66.5

74.8

58.2

Attendings (n \(=\) 15)

66.4

72.7

60.0

Senior physicians (n \(=\) 45)

67.7

73.0

62.3

Chief physicians (n \(=\) 3)

71.3

73.3

69.2

Practical experience (pe)

pe \(\leqslant \) 2 years (n \(=\) 46)

66.2

76.0

56.5

2 years < pe \(\leqslant \) 4 years (n \(=\) 37)

66.4

73.8

59.1

4 years < pe \(\leqslant \) 12 years (n \(=\) 32)

67.9

73.3

62.5

pe > 12 years (n \(=\) 42)

67.9

73.0

62.8