TABLE 3.
Classification performance across methods.
| Method | Laterality | Threshold | Sensitivity | Specificity | PPV | NPV | Accuracy |
|---|---|---|---|---|---|---|---|
| Optimal thresholds | Affected | ≥20 | 85.51 | 77.60 | 55.58 | 94.23 | 79.55 |
| Non-affected | ≥38 | 83.87 | 76.29 | 75.34 | 84.56 | 79.80 | |
| Bilateral | ≥20/≥38 | 76.08 | 82.33 | 50.66 | 93.52 | 81.13 | |
| Logistic regression | Affected | ≥0.5* | 78.56 | 79.89 | 56.15 | 91.91 | 79.56 |
| Non-affected | ≥0.5* | 79.46 | 75.05 | 72.26 | 81.71 | 77.04 | |
| Bilateral | ≥0.5/≥0.5* | 79.13 | 79.41 | 47.81 | 94.10 | 79.35 | |
| Conventional threshold | Affected | >2 | 94.42 | 63.68 | 46.01 | 97.21 | 71.27 |
| Non-affected | >2 | 95.03 | 57.56 | 65.92 | 93.06 | 74.93 | |
| Bilateral | >0 | 92.08 | 66.08 | 39.29 | 97.22 | 71.08 |
Legend: optimal thresholds on group level; ∗defold value of probability for logistic regression.