TABLE 4.
Case‐control classification accuracies and ROC AUC measures (on cross‐validation) with brain morphometric features in cMDD‐STR sample (30 cases and 30 controls)
| Classifier type | Feature selection | Hyperparam. optimisation | Outer CV | Inner CV | Feature domain | Classification accuracy (sensitivity/specificity) | ROC AUC |
|---|---|---|---|---|---|---|---|
| PLR | Embedded | Grid search | LOOCV | 10‐fold | Thickness | 60.00% (56.67/63.33%) | 0.609 |
| Surface area | 51.67% (50.00/53.33%) | 0.547 | |||||
| Volume | 56.67% (60.00/53.33%) | 0.539 | |||||
| Subcortical | 55.00% (56.67/53.33%) | 0.572 | |||||
| Combined | 50.00% (46.67/53.33%) | 0.546 | |||||
| SVM | None | None | LOOCV | ‐ | Thickness | 63.33% (63.33/63.33%) | 0.682 |
| Surface area | 46.67% (43.33/50.00%) | 0.529 | |||||
| Volume | 51.67% (40.00/63.33%) | 0.550 | |||||
| Subcortical | 60.00% (60.00/60.00%) | 0.582 | |||||
| Combined | 61.67% (70.00/53.33%) | 0.568 | |||||
| Grid search | LOOCV | Thickness | 60.00% (60.00/60.00%) | 0.556 | |||
| Surface area | 50.00% (46.67/53.33%) | 0.500 | |||||
| Volume | 61.67% (50.00/73.33%) | 0.649 | |||||
| Subcortical | 58.33% (50.00/66.67%) | 0.602 | |||||
| Combined | 58.33% (63.33/53.33%) | 0.628 | |||||
| Statistical filter | None | Combined | 53.33% (40.00/66.67%) | 0.540 | |||
| Grid search | 10‐fold | Combined | 45.00% (40.00/50.00%) | 0.513 | |||
| Sequential elimination | None | Thickness | 61.67% (56.67/66.67%) | 0.687 | |||
| Surface area | 48.33% (43.33/53.33%) | 0.519 | |||||
| Volume | 50.00% (36.67/63.33%) | 0.556 | |||||
| Subcortical | 53.33% (50.00/56.67%) | 0.573 | |||||
| Combined | 61.67% (63.33/60.00%) | 0.659 | |||||
| DT | None | None | LOOCV | ‐ | Thickness | 38.33% (40.00/36.67%) | 0.283 |
| Surface area | 75.00% (80.00/70.00%) | 0.680 | |||||
| Volume | 45.00% (43.33/46.67%) | 0.377 | |||||
| Subcortical | 43.33% (46.67/40.00%) | 0.394 | |||||
| Combined | 55.00% (50.00/60.00%) | 0.473 | |||||
| Grid search | LOOCV | Thickness | 51.67% (53.33/50.00%) | 0.318 | |||
| Surface area | 65.00% (76.67/53.33%) | 0.677 | |||||
| Volume | 46.67% (46.67/46.67%) | 0.442 | |||||
| Subcortical | 58.33% (56.67/60.00%) | 0.500 | |||||
| Combined | 38.33% (46.67/30.00%) | 0.407 | |||||
| Statistical filter | None | Combined | 33.33% (43.33/23.33%) | 0.189 | |||
| Grid search | 10‐fold | Combined | 43.33% (43.33/43.33%) | 0.350 | |||
| Sequential elimination | None | Thickness | 35.00% (46.67/23.33%) | 0.255 | |||
| Surface area | 68.33% (70.00/66.67%) | 0.637 | |||||
| Volume | 40.00% (40.00/40.00%) | 0.292 | |||||
| Subcortical | 51.67% (46.67/56.67%) | 0.426 | |||||
| Combined | 63.33% (56.67/70.00%) | 0.533 |
Note: Top accuracies for SVM, PLR and DT classifiers are in italics.
Abbreviations: CV, cross‐validation; DT, decision tree; LOOCV, leave‐one‐out cross‐validation; PLR, penalised logistic regression; ROC AUC, receiver operating characteristic area under the curve; SVM, support vector machine.