Table 3. Diagnostic performance of MLP-based IL and IWA models in the training, test, and external validation cohorts.
| Model | Method | AUC (95% CI) | Accuracy | Sensitivity | Specificity | PPV | NPV |
|---|---|---|---|---|---|---|---|
| IL (training) | MLP | 0.865 (0.804–0.916) | 0.807 | 0.904 | 0.587 | 0.832 | 0.730 |
| IL (test) | MLP | 0.833 (0.687–0.958) | 0.816 | 0.923 | 0.583 | 0.828 | 0.778 |
| IL (external) | MLP | 0.848 (0.685–0.965) | 0.774 | 0.773 | 0.778 | 0.895 | 0.583 |
| IWA (training) | MLP | 0.927 (0.875–0.966) | 0.847 | 0.923 | 0.674 | 0.865 | 0.795 |
| IWA (test) | MLP | 0.872 (0.716–0.992) | 0.842 | 0.885 | 0.750 | 0.885 | 0.750 |
| IWA (external) | MLP | 0.859 (0.603–1.000) | 0.806 | 0.864 | 0.667 | 0.864 | 0.667 |
AUC, area under the curve; CI, confidence interval; IL, infarct lesion; IWA, combination of infarct lesion, affected hemisphere, and whole brain; MLP, multilayer perceptron; NPV, negative predictive value; PPV, positive predictive value.