Table 4.
The diagnostic performance of radiologist assessment, LR model, and LR-SVMSMOTE model in the prediction of thyroid cartilage invasion
| Radiologist | LR | LR-SVMSMOTE | |
|---|---|---|---|
| AUC(95%CI) | 0.721(0.663–0.774) | 0.876(0.830–0.913) | 0.905(0.863–0.937) |
| Sensitivity (%) | 74.4 | 80.2 | 80.2 |
| Specificity (%) | 69.8 | 83.8 | 88.3 |
| Accuracy (%) | 71.3 | 82.6 | 85.7 |
| Precision(%) | 0.542 | 0.704 | 0.767 |
| F1-score | 0.627 | 0.750 | 0.784 |
| Kappa | 0.404 | 0.618 | 0.677 |
| MCC | 0.417 | 0.621 | 0.677 |
| P Value | < 0.001a | < 0.001a |
LR Logistic regression, LR-SVMSMOTE Logistic regression - support vector machine-based synthetic minority oversampling, CI Confidence Interval, MCC Matthews Correlation Coefficient
aDelong test for differences in AUC compared to radiologist assessment