Table 1.
Characteristics of each of the twelve studies included in the systematic review.
| Study | Cohort | Nr. of subjects (mean age, nr. of females) | Neuroimaging acquisition | Features of interest | Classification parameters | Performance validation | Results |
|---|---|---|---|---|---|---|---|
| Tripoliti et al., 2007 | AD HC |
12 (77.2, 7) 14 (74.9, 9) |
– 1.5 T MRI – Task-based fMRI |
– Demographic data; – Behavioral data; – Head motions parameters; – Volumetric measures; – Activation patterns; – BOLD-derived hemodynamic measures. |
– Feature selection based on correlation; – RF with 10 trees. |
10-fold cross-validation sensitivity/specificity | AD vs. HC: 98%/98% |
| Gray et al., 2013 | AD sMCI pMCI HC |
37 (76.8, 14) 34 (75.7, 12) 41 (76.1, 12) 35 (74.5, 12) |
– 1.5 T MRI – FDG-PET |
– Volumetric measures; – FDG-PET voxel intensities whole-brain; – CSF-derived measures; – Genetic information. |
– RF with 5,000 trees. | Stratified repeated random sampling accuracy on a separate test set | AD vs. HC: 89% MCI vs. HC: 74.6% sMCI vs. pMCI: 58.4% |
| Cabral et al., 2013 | AD MCI HC |
59 (78.2, 25) 59 (77.7, 19) 59 (77.4, 21) |
– FDG-PET | – FDG-PET voxel intensities; | – Feature selection with Mutual Information criterion; – Decomposition by the one-vs.-all scheme; – Aggregation scheme with voting strategy (MAX); – RF with 100 trees. |
Repeated 10-fold cross-validation accuracy | AD vs. MCI vs. HC: 64.63% |
| Lebedev et al., 2014 | AD HC |
185 (75.2, 92) 225 (75.95, 110) |
– 1.5 T MRI | – Non-cortical volumes; – Cortical thickness; Jacobian maps; – Sulcal depth. |
– Recursive feature elimination with Gini index; – RF with 1,000 trees. |
Overall accuracy on a separate test set | AD vs. HC: 90.3% |
| Moradi et al., 2015 | AD sMCI pMCI HC |
200(55-91, 97) 100 (57-89, 34) 164 (77-89, 67) 231 (59-90, 112) |
– 1.5 T MRI | – GM density values; – Age; – Neuropsychological scores. |
– Feature selection with regularized logistic regression framework | 10-fold cross-validation accuracy | sMCI vs. pMCI: 82% |
| Oppedal et al., 2015 | AD LBD HC |
57 (N.A.) 16 (N.A.) 36 (N.A.) |
– 1.0/1.5 T MRI– FLAIR | – Local binary pattern (LBP); – Image contrast measure (C). |
– RF with 10 trees. | 10-fold nested cross-validation accuracy | AD vs. LBD vs. HC: 87% AD+LBD vs. HC: 98% AD vs. LBD: 74% |
| Sivapriya et al., 2015 | AD MCI HC |
140 (N.A.) 450 (N.A.) 280 (N.A.) |
– MRI – FDG-PET |
– Volumetric measures; – FDG-PET uptake ROI-based; – Neuropsychological scores. |
– Feature selection with particle swarm optimization approach coupled with the Merit Merge technique (CPEMM); – RF with 100 to 1,000 trees. |
5-fold cross-validation accuracy | AD vs. MCI vs. HC: 96.3% |
| Wang et al., 2016 | sMCI pMCI |
65 (72.2, 26) 64 (72.5, 29) |
– 1.5 T MRI – florbetapir-PET – FDG-PET |
– Morphological measures; – florbetapir-PET uptake whole-brain; – FDG-PET uptake whole-brain. |
– RF with 500 trees. | Leave-one-out cross-validation accuracy | sMCI vs. pMCI: 73.64% |
| Ardekani et al., 2017 | sMCI pMCI |
78 (74.75, 24) 86 (74.10, 31) |
– 1.5 T MRI | – Hippocampal volumetric integrity; – Neuropsychological scores. |
– Feature selection with Gini index; – RF with 5,000 trees. |
OOB estimation of classification accuracy | sMCI vs. pMCI: 82.3% |
| Lebedeva et al., 2017 | MCI HC |
32 (78.1, 22) 40 (76.4, 29) |
– 1.5/3 T MRI | – Cortical thickness; – Subcortical volumes. – MMSE |
– Feature selection with Gini index; – RF with 5,000 trees. |
OOB estimation of classification accuracy | MCI vs. HC: 81.3% |
| Maggipinto et al., 2017 | AD MCI HC |
50 (N.A.) 50 (N.A.) 50 (N.A.) |
– DTI | – TBSS FA voxels; | – Feature selection with the Wilcoxon rank sum test and ReliefF algorithm; – RF with 300 trees. |
Repeated 5-fold cross-validation accuracy | AD vs. HC: 87% MCI vs. HC: 81% |
| Son et al., 2017 | AD MCI HC |
30 (74, 18) 40 (74.3, 21) 35 (76.06, 23) |
– 3 T MRI – rs-fMRI |
– Subcortical volumes; – Eigenvector centrality of functional networks ROI-based. |
N.A. | Repeated leave-one-out cross-validation accuracy | AD vs. MCI vs. HC: 53.33% |
Data are related to the highest performance reached by random forest. AD, Alzheimer's disease; HC, healthy controls; MCI, Mild cognitive impairment; cMCI, converter MCI; pMCI, progressive MCI; LBD, Lewy-body dementia; MRI, Magnetic resonance imaging; fMRI, functional MRI; rs-fMRI, resting state fMRI; PET, positron emission tomography; FDT-PET, fluorodeoxyglucose PET; DTI, Diffusion tensor imaging; GM, Gray matter; ROI, Region of interest; MMSE, Mini mental state examination; TBSS, Tract-based spatial statistics; OOB, out-of-bag; N.A., not applicable.