Table 1.
Characteristics of included studies and heterogeneity factors.
| Study (year) | Diagnostic task | MRI modality | Data source | Validation strategy | Algorithms (comparison) | Reported accuraciesa | Reported specificitiesa | Reported sensitivitiesa | Participantsb |
|---|---|---|---|---|---|---|---|---|---|
| Plant et al. (2010) | Diagnosis classification and prediction of MCI conversion | T1-weighted | Local cohort | NR | Bayes vs. SVM | Bayes: 92%; SVM: 98% | Bayes: 89%; SVM: 100% | Bayes: 94%; SVM: 96% | HC 18; MCI 24; AD 32 |
| Escudero et al. (2011) | AD vs. healthy control classification | T1-weighted | ADNI | NR | LR vs. SVM | LR: 86%; SVM: 89% | Not reported | Not reported | HC 180; MCI 222; AD 122 |
| Diciotti et al. (2012) | AD vs. healthy control classification | T1-weighted | Local cohort | NR | Bayes vs. SVM | Bayes: 78%; SVM: 86% | Bayes: 83%; SVM: 90% | Bayes: 72%; SVM: 82% | HC 29; MCI 30; AD 21 |
| Aguilar et al. (2013) | Diagnosis classification and prediction of MCI conversion | T1 + T2 | AddNeuroMed (public) | NR | MP vs. SVM | MP: 87%; SVM: 88% | MP: 90%; SVM: 90% | MP: 85%; SVM: 86% | HC 110; MCI 119; AD 116 |
| Moradi et al. (2015) | Diagnosis classification and prediction of MCI conversion | T1-weighted | ADNI | NR | LDS vs. SVM | LDS: 75%; SVM: 82% | LDS: 52%; SVM: 74% | LDS: 89%; SVM: 87% | HC 231; MCI 394; AD 200 |
| Lama et al. (2017) | Diagnosis classification and prediction of MCI conversion | T1-weighted | ADNI | NR | ELM vs. SVM | ELM: 78%; SVM: 80% | ELM: 84%; SVM: 79% | ELM: 69%; SVM: 83% | HC 70; MCI 74; AD 70 |
| Cohen et al. (2019) | AD vs. healthy control classification | T1-weighted | ADNI | NR | FFR vs. NN | FFR: 87%; NN: 88% | Not reported | Not reported | HC 352; MCI 531; AD 334 |
| Pan et al. (2020) | AD vs. healthy control classification | T1-weighted | ADNI | NR | CNN vs. SVM | CNN: 84%; SVM: 76% | Not reported | Not reported | HC 162; MCI 210; AD 137 |
| Liu et al. (2020) | AD vs. healthy control classification | T1-weighted | ADNI | NR | DenseNet vs. CNN | DenseNet: 83%; CNN: 76% | DenseNet: 76%; CNN: 76% | DenseNet: 89%; CNN: 77% | HC 112; MCI 180; AD 97 |
| Zhang J. et al. (2021) | AD vs. healthy control classification | T1-weighted | ADNI | NR | DenseNet vs. ResNet vs. CNN | DenseNet: 95%; ResNet: 93%; CNN: 85% | DenseNet: 95%; ResNet: 93%; CNN: 88% | DenseNet: 94%; ResNet: 94%; CNN: 83% | HC 275; MCI 413; AD 280 |
a
Accuracies, specificities and sensitivities are expressed as percentages. “NR” indicates the metric was not available in the original study.
b
Participant counts refer to the number of healthy controls (HC), individuals with mild cognitive impairment (MCI) and patients with Alzheimer’s disease (AD) included in each study.