Table 1.
Number of comparisons in each systematic review analysis group using specified data source, machine learning method, types of imaging and nonimaging data, and by study size
| Data sources | HC versus AD | HC versus MCI | MCInc versus MCIc | MCI versus AD | Total |
|---|---|---|---|---|---|
| ADNI | 54 | 24 | 34 | 7 | 119 |
| ADNI + Bdx-3C | 0 | 0 | 1 | 0 | 1 |
| AddNeuroMed | 1 | 0 | 2 | 0 | 3 |
| AddNeuroMed + ADNI | 2 | 1 | 1 | 0 | 4 |
| Local | 4 | 3 | 0 | 0 | 7 |
| OASIS | 7 | 2 | 0 | 1 | 10 |
| Total | 68∗ | 30 | 38 | 8 | 144 |
| Machine learning method | |||||
| AdaBoost | 1 | 0 | 1 | 0 | 2 |
| Deep Learning | 2 | 2 | 0 | 0 | 4 |
| Gaussian process | 0 | 0 | 1 | 0 | 1 |
| LDA | 5 | 0 | 5 | 1 | 11 |
| Logistic regression | 4 | 0 | 2 | 0 | 6 |
| OPLS | 2 | 1 | 1 | 0 | 4 |
| QDA | 0 | 0 | 1 | 0 | 1 |
| RBF-NN | 0 | 0 | 1 | 0 | 1 |
| Random forest | 3 | 1 | 3 | 0 | 7 |
| SRC | 2 | 1 | 2 | 0 | 5 |
| SVM | 39 | 22 | 17 | 7 | 85 |
| SVM + MKL | 3 | 1 | 1 | 0 | 5 |
| SVM + OPLS | 1 | 0 | 1 | 0 | 2 |
| SVM + random forest | 2 | 1 | 2 | 0 | 5 |
| SVM + SRC | 1 | 1 | 0 | 0 | 2 |
| kNN | 3 | 0 | 0 | 0 | 3 |
| Total | 68∗ | 30 | 38 | 8 | 144 |
| Types of imaging and imaging plus nonimaging data used | |||||
| T1w only | 46 | 13 | 26 | 6 | 91 |
| T1w and other imaging data | 8 | 8 | 2 | 0 | 18 |
| T1w and other types of data | 8 | 3 | 8 | 1 | 20 |
| T1w and both other imaging and types of data | 6 | 6 | 2 | 1 | 15 |
| Total | 68∗ | 30 | 38 | 8 | 144 |
| Size of data set (range from 100 to 902 participants) | |||||
| 150 and under | 30 | 4 | 9 | 2 | 45 |
| 151 to 200 | 4 | 10 | 6 | 0 | 20 |
| 201 to 250 | 9 | 4 | 6 | 0 | 19 |
| 251 to 300 | 4 | 2 | 3 | 0 | 9 |
| Over 300 | 21 | 10 | 14 | 6 | 51 |
| Total | 68∗ | 30 | 38 | 8 | 144 |
Abbreviations: HC, healthy control; AD, Alzheimer's disease; MCI, mild cognitive impairment; nc, nonconverter to AD; T1w, T1-weighted magnetic resonance imaging; c, converter to AD; LDA, linear discriminant analysis; KNN, k-nearest neighbors; OPLS, Orthogonal Projections to Latent Structures; SRC, Sparse Representation Classification.
NOTE. Individual studies contribute to more than one analysis and use more than one data source, machine learning method, combinations of imaging data, and more than one data set (hence more than one sample size in some studies).
In the 68 HC versus AD comparisons, one study is counted twice as it used two different kinds of imaging.