Table 3.
Comparison with related studies
| Study | Modality | Accuracy | F1-score | Method |
|---|---|---|---|---|
| Bucholc et al, 201918 | MRI, PET, clinical | 82.90% | Not reported | SVM |
| Fang et al 202026 | MRI, PET | 66.29% | Not reported | GDCA |
| Abuhmed et al, 202117 | MRI, PET, clinical | 86.08% | 87.67% | Multivariate BiLSTM |
| Venugopalan et al, 202119 | MRI, SNP, clinical | 78% | 78% | DL + RF |
| MADDi | MRI, SNP, clinical | 96.88% | 91.41% | DL + attention |
Note: This table shows the comparison between our study and 5 other previous studies that attempted to solve a similar problem to ours. MADDi performed with 96.88% average accuracy and 91.41% average F1-score across 5 random initializations on a held-out test set, achieving state-of-the-art performance on the multimodal 3-class classification task.