TABLE 3.
Summary of evaluation metrics for the ensemble model when varying the transfer learning module composing the baseline models, either ViT or CNN architectures. For each metric, bold values indicate the optimal achieved values.
| Ensemble model | ||||
|---|---|---|---|---|
| ViT | Xception | Densenet201 | ResNet101 | |
| AUC (%) | 91.4 | 78.6 | 62.9 | 57.1 |
| Accuracy (%) | 82.4 | 70.6 | 64.7 | 64.7 |
| Specificity (%) | 80.0 | 90.0 | 57.1 | 60.0 |
| Sensitivity (%) | 85.7 | 42.9 | 70.0 | 71.4 |
| Precision (%) | 75.0 | 75.0 | 57.1 | 55.6 |
| F‐score (%) | 80.0 | 54.6 | 57.1 | 62.5 |
| G‐mean (%) | 82.8 | 62.1 | 63.3 | 65.5 |