Table 2.
Comparison with previous studies for EMCI/LMCI classification.
| Authors | Target | Approach | Feature extraction | Feature selection | Classifier | ACC | AUC |
|---|---|---|---|---|---|---|---|
| Goryawala et al. (2015) | EMCI (114) vs. LMCI(91) | MRI (Cortical volume) + Neuropsychological scores | *SLRM for MRI and Neuropsychological test | *LDA | 0.736 | N/A | |
| Jie et al. (2018) | EMCI (56) vs. LMCI(43) | DCN(dynamic connect. network) from rs-fMRI | Temporal & spatial Variability (DCNs) | *M2FL method | SVM | 0.788 | 0.783 |
| Nozadi et al. (2018) | EMCI(164) vs. LMCI(189) | FDG-PET, AV45-PET | Multimodal PET-MRI registration + ROIs-based or whole brain select | RF | 0.725 | 0.79 | |
| Sheng et al. (2019) | EMCI(24) vs. LMCI(24) | BCT(Brain Connectivity Toolbox) from rs-fMRI | Network-based measures (BCT) | mRMR, Chi-square, Gini score,Kruskal–Wallis test, Fisher score(FS), Relief feature score | > 20Classifiers + *DNN | 0.875 (SVM) | N/A |
| Zhang et al. (2019) | EMCI(33) vs. LMCI(29) | Graph theory (rs-fMRI) | 3Network features + 3 freq. bands | mRMR, SS-LR, Fisher Score (FS) | SVM | 0.838 | 0.905 |
| Shi and Liu (2020) | EMCI(77) vs. LMCI(64) | rs-fMRI | Hilbert-Huang transform (HHT) Hilbert weighted frequencies(HWFs) Independent two-sample t-test | SVM | 0.879 | N/A | |
*SLRM stepwise linear regression models.
*M2FL method Manifold regularized multi- task feature selection.
*DCN dynamic connectivity network.
*LDA linear discriminant analysis.
*DNN deep neural network.