Table 4 Meta-analysis results, summary estimates of sensitivity, specificity, and likelihood ratio according to subgroups

From: A systematic review and meta-analysis of artificial intelligence versus clinicians for skin cancer diagnosis

 

Measure

Sensitivity

Specificity

LR +

LR -

Overall clinicians (n = 19 studies)

Summary estimate AI

87.0% (95% CI 81.7–90.9%)

77.1% (95% CI 69.8–83.0%)

3.79 (95% CI 2.89–4.97)

0.17 (95% CI 0.12–0.23)

Summary estimate overall clinicians

79.8% (95% CI 73.2–85.1%)

73.6% (95% CI 66.5–79.6%)

3.02 (95% CI 2.33 – 3.91)

0.27 (95% CI 0.20–0.37)

Generalists (n = 5 studies)

Summary estimate AI

92.5% (95% CI 88.9–94.9%)

66.5% (95% CI 56.7–75.0%)

2.76 (95% CI 2.10-3.61)

0.11 (95% CI 0.07-0.16)

Generalist

64.6% (95% CI 47.1–78.9%)

72.8% (95% CI 56.7–84.5%)

2.37 (95% CI 1.63-3.46)

0.48 (95% CI 0.34–0.69)

Non-expert dermatologist (n = 14 studies)

Summary estimate AI

85.4% (95% CI 78.9–90.2%)

78.5% (95% CI 70.6–84.8%)

3.98 (95% CI 2.89–5.49)

0.18 (95% CI 0.13–0.27)

Non-experts

76.4% (95% CI 71.1–80.9%)

67.1% (95% CI 57.2–75.6%)

2.32 (95% CI 1.71–3.14)

0.35 (95% CI 0.27-0.46)

Expert dermatologist (n = 16 studies)

Summary estimate AI

86.3% (95% CI 80.4–90.7%)

78.4% (95% CI 71.1–84.3%)

3.99 (95% CI 2.97-5.37)

0.17 (95% CI 0.12–0.25)

Experts

84.2% (95% CI 76.2–89.8%)

74.4% (95% CI 65.3–81.8%)

3.29 (95% CI 2.31-4.67)

0.21 (95% CI 0.13–0.34)

  1. Abbreviations: LR + = positive likelihood ratio; LR−= negative likelihood ratio.
Back to article page