Table 7.
Comparison of different fusion strategies on the external Sherif dataset.
| Fusion strategy | Accuracy (%) | Macro-F1 (%) | MCC |
|---|---|---|---|
| Best CNN branch | 93.50 | 90.10 | 0.905 |
| Best ViT branch | 92.80 | 89.50 | 0.890 |
| Softmax averaging | 93.10 | 89.90 | 0.895 |
| Weighted averaging | 93.40 | 90.00 | 0.900 |
| Logistic regression stacking | 93.80 | 90.40 | 0.907 |
| AAS (CNN + ViT) | 94.00 | 90.80 | 0.910 |
| AAS (CNN + ViT + Radiomics) | 94.00 | 91.00 | 0.915 |
Bold values highlight the proposed NeuroTrustNet/AAS configuration.