Performance of binary classification across all models and experts in the test set
| Method | AUC | Accuracy (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) |
|---|---|---|---|---|
| Radiomics (by TPOT) | ||||
| MB vs non-MB | 0.94 | 0.85 (0.74–0.92) | 0.91 (0.72–0.99) | 0.81 (0.67–0.90) |
| EP vs non-EP | 0.84 | 0.80 (0.69–0.88) | 0.52 (0.32–0.71) | 0.93 (0.81–0.98) |
| PA vs non-PA | 0.94 | 0.88 (0.78–0.94) | 0.95 (0.76–1.00) | 0.84 (0.70–0.92) |
| Radiomics (by manual optimized pipeline) | ||||
| MB vs non-MB | 0.98 | 0.91 (0.81–0.96) | 0.96 (0.78–1.00) | 0.88 (0.75–0.95) |
| EP vs non-EP | 0.70 | 0.71 (0.59–0.81) | 0.19 (0.07–0.40) | 0.95 (0.83–0.99) |
| PA vs non-PA | 0.93 | 0.86 (0.75–0.93) | 0.77 (0.56–0.90) | 0.91 (0.78–0.97) |
| Expert 1 | ||||
| MB vs non-MB | NA | 0.67 (0.55–0.77) | 0.65 (0.45–0.81) | 0.67 (0.52–0.79) |
| EP vs non-EP | NA | 0.74 (0.60–0.82) | 0.57 (0.36–0.75) | 0.82 (0.68–0.91) |
| PA vs non-PA | NA | 0.74 (0.62–0.83) | 0.50 (0.31–0.69) | 0.86 (0.72–0.94) |
| Expert 2 | ||||
| MB vs non-MB | NA | 0.64 (0.52–0.75) | 0.57 (0.37–0.75) | 0.66 (0.51–0.77) |
| EP vs non-EP | NA | 0.68 (0.54–0.79) | 0.43 (0.25–0.64) | 0.80 (0.66–0.89) |
| PA vs non-PA | NA | 0.68 (0.56–0.78) | 0.50 (0.31–0.69) | 0.77 (0.63–0.87) |