TABLE II.
SHARING WEIGHTS IMPACT ON THE PERFORMANCE THE PROPOSED ATTENTION-GUIDED MODEL
| Accuracy | MCC | Recall | Precision | F1-score | AUC | |
|---|---|---|---|---|---|---|
| No weight sharing: each pathway learns its weights | ||||||
| Pathway#1 | 90.349 | 0.496 | 94.311 | 94.880 | 94.595 | 90.198 |
| Pathway#2 | 91.413 | 0.553 | 94.838 | 95.418 | 95.102 | 93.007 |
| Pathway#3 | 92.225 | 0.580 | 95.808 | 95.522 | 95.665 | 88.579 |
| Fusion - average | 91.421 | 0.521 | 95.808 | 94.675 | 95.238 | 91.392 |
| Fusion - min | 92.225 | 0.484 | 99.701 | 92.244 | 95.827 | 90.095 |
| Fusion - max | 89.008 | 0.533 | 91.018 | 96.508 | 93.683 | 90.922 |
| Full weight sharing: Layers of all pathways share weights | ||||||
| Pathway#1 | 90.349 | 0.585 | 91.916 | 97.152 | 94.462 | 90.527 |
| Pathway#2 | 91.153 | 0.586 | 93.413 | 96.594 | 94.977 | 91.528 |
| Pathway#3 | 92.493 | 0.599 | 95.808 | 95.808 | 95.808 | 91.475 |
| Fusion - average | 91.689 | 0.611 | 93.713 | 96.904 | 95.282 | 92.210 |
| Fusion - min | 93.298 | 0.601 | 97.904 | 94.783 | 96.318 | 91.481 |
| Fusion - max | 89.008 | 0.588 | 89.521 | 98.033 | 93.584 | 91.137 |
| Partial sharing: Only pathway 1 and 2 share weights | ||||||
| Pathway#1 | 89.544 | 0.520 | 92.216 | 95.950 | 94.046 | 90.970 |
| Pathway#2 | 90.080 | 0.534 | 92.814 | 95.975 | 94.368 | 93.054 |
| Pathway#3 | 92.225 | 0.590 | 95.509 | 95.796 | 95.652 | 93.432 |
| Fusion - average | 90.349 | 0.553 | 92.814 | 96.273 | 94.512 | 94.405 |
| Fusion - min | 93.298 | 0.584 | 98.802 | 94.017 | 96.350 | 92.647 |
| Fusion - max | 88.204 | 0.559 | 88.922 | 97.697 | 93.103 | 94.505 |
| FusionAvgPath.1&2 | 89.812 | 0.539 | 92.216 | 96.250 | 94.190 | 92.930 |
| FusionAvgPath.1&3 | 92.493 | 0.627 | 94.910 | 96.646 | 95.770 | 94.528 |
| FusionAvgPath.2&3 | 94.102 | 0.699 | 96.108 | 97.273 | 96.687 | 94.271 |