Table 5.
Comparison of the proposed method with other traditional methods and deep-learningbased methods in sMCI/pMCI classification on ADNI dataset.
| Method | Type | Modalities | sMCI/pMCI | BACC |
|---|---|---|---|---|
| Cross-sectional | ||||
| Liu et al. (2018) | D | MRI | 465/205 | 62.2 |
| Zu et al. (2016) | N | MRI, PET | 56/43 | 69.0 |
| Suk et al. (2014) | N+D | MRI | 128/76 | 63.8 |
| Lin et al. (2018) | D | MRI | 100/164 | 73.0* |
| Huang et al. (2019) | D | MRI, PET | 441/326 | 76.9 |
| Zhou et al. (2019a) | N | MRI, PET, SNP | 205/157 | 74.3* |
| Zhou et al. (2019b) | N | MRI, PET | 114/71 | 78.3 |
| Zeng et al. (2021) | D | MRI, clinical measures | 82/95 | 87.8* |
| Nguyen et al. (2021) | D | MRI | 129/171 | 74.0 |
| Yuan et al. (2021) | N | MRI, SNP | 115/113 | 82.4 |
| Shen et al. (2021) | N | MRI | 59/55 | 65.7 |
| Longitudinal | ||||
| Gray et al. (2012) | N | MRI, PET | 64/53 | 62.7 |
| Cui and Liu (2019) | D | MRI | 236/167 | 71.7 |
| Platero and Tobar (2020) | N | MRI, clinical measures | 215/206 | 77.1 |
| Ours | D | MRI | 193/135 | 73.5 |
‘D’ denotes deep-learning methods, and ‘N’ denotes non-deep-learning methods.
*
Refers to ACC scores, i.e., classification accuracy not accounting for imbalance between cohort sizes. The proposed method achieved the second-highest accuracy among all methods that were solely based on structural MRI.