Table 2.
AD/HC performance of the final models: ADNI cohort.
| Models | OOB error [Train] | Sensitivity/specificity (OA) [Test] | AUC (95% C.I.) | |
|---|---|---|---|---|
| Cortical thickness | RFE | 14.0% | 88.6%/90.7% (89.62%) | 0.93 (0.9–0.96) |
| Sulcal depth | 21.3% | 80.0%/74.7% (77.3%) | 0.84 (0.8–0.89) | |
| Jacobian | 21.7% | 77.1%/81.3% (79.2%) | 0.84 (0.79–0.88) | |
| Volumes | 15.0% | 82.9%/86.7% (84.7%) | 0.91 (0.88–0.95) | |
| Thickness + volumes (ROI) | 11.7% | 88.6%/92.0% (90.3%) | 0.94 (0.91–0.96) | |
| Thickness | no RFE | 14.7% | 88.6%/89.3% (89.0%) | 0.92 (0.89–0.95) |
| Sulcal depth | 14.0% | 80.0%/73.3% (76.67%) | 0.83 (0.79–0.88) | |
| Jacobian | 21.0% | 80.0%/80.0% (80.0%) | 0.84 (0.79–0.88) | |
| Volumes | 16.7% | 80.0%/86.7% (83.3%) | 0.91 (0.88–0.94) | |
| Thickness + volumes (ROI) | 12% | 85.7%/89.3% (87.5%) | 0.93 (0.91–0.96) | |
AD/HC — Alzheimer's disease/healthy controls; OA — overall accuracy; OOB — out-of-bag estimate; AUC (95% C.I.) — area under the ROC curve with 95% confidence interval.