Table 3 Diagnostic performance by different models
Model | AUC (95% CI) | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | Accuracy (%) |
|---|---|---|---|---|---|---|
Training set (n = 72) | ||||||
 Clinical model | 0.761 (0.946–0.871) | 73.7 | 77.4 | 53.8 | 89.1 | 76.4 |
 Radiomics model | 0.948 (0.909–0.987) | 83.7 | 88.4 | 89.1 | 82.6 | 85.9 |
  Combined model | 0.966(0.932–0.999) | 87.5 | 90.9 | 91.3 | 87.0 | 89.1 |
  Reader A | 0.804(0.723–0.886) | 82.6 | 78.3 | 79.2 | 81.8 | 80.4 |
  Reader B | 0.739(0.658–0.820) | 93.5 | 54.3 | 67.2 | 89.3 | 73.9 |
  Reader C | 0.674(0.588–0.760) | 89.1 | 45.7 | 62.1 | 80.8 | 67.4 |
Test set (n = 18) | ||||||
 Clinical model | 0.766 (0.525–1) | 80.0 | 76.9 | 57.1 | 90.9 | 77.8 |
 Radiomics model | 0.909 (0.770–1) | 80.0 | 76.9 | 57.1 | 90.9 | 77.8 |
 Combined model | 0.896 (0.733–1) | 80.0 | 76.9 | 57.1 | 90.9 | 77.8 |
  Reader A | 0.740(0.523–0.957) | 57.1 | 90.9 | 80.0 | 76.9 | 77.8 |
  Reader B | 0.604(0.356–0.852) | 57.1 | 63.6 | 50.0 | 70.0 | 61.1 |
  Reader C | 0.513(0.262–0.764) | 57.1 | 45.4 | 40.0 | 62.5 | 50.0 |