Table 4.
Datasets, data analysis methods, and key findings for selected studies in research foci 2 - predicting MCI to AD conversion.
| Study | Datasets | Primary data analysis methods | Key findings |
|---|---|---|---|
| Chen et al. (35) | MRI and magnetic resonance spectroscopy data (8 PMCI, 18 SMCI), ADNI (for validation) | Bayesian network (Bayesian Outcome prediction with Ensemble Learning (BOPEL)) based on brain volums in different regions | Bayesian data mining with ensemble learning demonstrates high predictive accuracy for MCI to AD conversion. |
| Gracia- Garcia et al. (34,36) | 3,864 participants from a longitudinal epidemiological study Zaragoza Dementia and Depression Project (ZARADEMP) | Multivariate model | Severe depression increases the risk of AD |
| Liu et al. (37) | MRI data from ADNI (86 AD, 93 SMCI, 97 MCI converters, 137 HC) | Locally linear embedding (LLE) by transforming multivariate MRI data to a locally linear space | LLE can improve the performance for predicting MCI to AD conversion |
| Hinrichs et al. (38) | ADNI subjects with MRI and FDG-PET scans (48 AD, 119 MCI, 66HC) | Support vector machine, Multi-kernel learning | MKL outperforms SVM. Imaging modalities are better predictors than neuropsychological scores. |
| Challis et al. (39) | Clinical measurements and MRI scans from 27 AD, 50 MCI and 39 HC | Bayesian Gaussian process logistic regression (GP-LR) | GP-LR can accurately differentiate MCI versus HC and MCI versus AD. |
| Li et al. (40) | MRI scans from ADNI database (180 AD, 160 MCI converters, 214 MCI non-converters, 204 HC) | Deep learning | Deep learning can distinguish four stages of AD progression using MRI with less clinical prior knowledge. |
| Lovestone et al. (41,42) | Behavioral assessment, hippocampal morphology, MRI in AddNeuroMed study (71 AD, 103 MCI, 88 HC) | Support vector machine | Hippocampal shape analysis provides a prognostic biomarker to predict MCI to AD conversion |
| Poil et al. (43) | MRI, EEG and clinical data (25 PMCI, 39 SMCI) from subjects referred to Alzheimer Center in Netherland | Elastic net logistic regression | Six EEG biomarkers mainly related to activity in the beta- frequency range (13–30 Hz) can predict conversion from MCI to AD (sensitivity of 88%, specificity of 82%) |
| Pozueta et al. (44) | Clinical and neuropsychological evaluation (55 SMCI, 50 PMCI) | Logistic regression | A combination of MMSE and an episodic memory test can predict MCI to AD conversion. (PPV: 80.95%) |
| Zhang et al. (32) | MRI, FDG-PET, CSF data from ADNI (45 AD, 91 MCI, 50 HC) | Multi-modal multi-task (M3T) learning | M3T learning performed well on both AD detection and clinical score prediction |
| Gomar et al. (46) | Clinical, cognitive, MRI, positron emission tomography, and cerebrospinal fluid from ADNI1 (150 PMCI, 168 SMCI) | Logistic regression | Cognitive measures especially an episodic memory measure and clock drawing screening test were evaluated to be great predictors for MCI to AD conversion. |
| Alegret et al. (47) | Neuropsychological measurements, brain SPECT data from 42 AD, 42 MCI (25 PMCI, 14 SMCI), and 42 HC | Correlation analysis, Cox regression analysis | Extent of memory impairment associated with speed of MCI to AD conversion |
| Runtti et al. (48) | Neuropsychological tests and AD biomarkers in ADNI (140 PMCI, 149 SMCI) | Machine learning based disease state index (DSI) method, linear regression | DSI to quantify longitudinal clinical data can predict MCI to AD conversion. (76.9% accuracy) |
Note: ADNI, Alzheimer’s Disease Neuroimaging Initiative database; MCI, mild-cognitive impairment; SMCI, stable mild cognitive impairment; PMCI, progressive mild cognitive impairment; HC, healthy control.