Table 5.
Comparison between the proposed JSRL method and the state-of-the-art neuroimaging studies in the task of MCI conversion prediction (i.e., pMCI vs. sMCI classification) on the ADNI database. N/P (training): Negative/Positive samples in the training set; N/P (test): Negative/Positive samples in the test set.
| Method | N/P (training) | N/P (test) | Modality | AUC | ACC |
|---|---|---|---|---|---|
| Shi et al. (2017) | 102/89 | 6/5 | MRI+PET | 0.801 | 0.789 |
| Suk etal. (2014) | 115/68 | 13/8 | MRI+PET | 0.747 | 0.759 |
| Young et al. (2013) | 36/35 | 42/30 | MRI+PET+APOE+CSF | 0.763 | 0.722 |
| Lian et al. (2020) | 455/197 | 239/38 | MRI | 0.781 | 0.809 |
| Pan et al. (2020) | 477/308 | 256/89 | MRI+PET | 0.825 | 0.778 |
| JSRL (Ours) | 530/409 | 66/93 | MRI+PET | 0.838 | 0.780 |