Discriminant Models
| NA vs. (converting MCI+AD) | Nonconverting MCI vs. (converting MCI+AD) | |||||||
| 1 component | 4 components | 1 component | 4 components | |||||
| Parameter | Exp. | CI | Exp. | CI | Exp. | CI | Exp. | CI |
| Model performance | ||||||||
| Sensitivity | 75.8 | 70.1–82.7 | 90.1 | 85.9–95.3 | 81.9 | 75.7–88.6 | 83.2 | 77.2–89.2 |
| Specificity | 83.3 | 72.1–94.6 | 88.1 | 78.3–97.9 | 77.8 | 62.1–93.5 | 85.2 | 71.8–98.6 |
| Accuracy | 77.5 | 71.6–83.4 | 90.0 | 85.8–94.3 | 81.3 | 75.5–87.0 | 83.5 | 78.0–89.0 |
| ROC AUC | 85.5 | 79.3–90.6 | 93.1 | 88.0–95.7 | 87.2 | 80.4–92.7 | 89.4 | 83.3–93.3 |
| Within-group classification | NA | AD | NA | AD | NC | AD | NC | AD |
| NA | 81.0 | 19.0 | 92.9 | 7.1 | 81.0* | 19.0* | 88.1* | 11.9* |
| Nonconverters | 66.7* | 33.3* | 88.2* | 14.8* | 85.2 | 14.8 | 96.3 | 3.7 |
| Early MCI | 29.7 | 70.3 | 18.9 | 81.1 | 35.1 | 64.9 | 29.7 | 70.3 |
| Late MCI | 27.6 | 72.4 | 19.0 | 81.0 | 29.3 | 70.7 | 20.7 | 79.3 |
| AD | 13.0 | 87.0 | 14.8 | 85.2 | 13.0 | 87.0 | 14.8 | 85.2 |
* Not involved in training step.
Exp. = expected value; CI = confidence interval; ROC AUC = area under receiver-operating-characteristic curve; NA = normal aging.
Discriminant models are as evaluated by leave-one-out cross-validation considering partitions into two contrasting groups: NA vs. all AD and nonconverting MCI vs. all AD. Linear discrimination was applied to best discriminant region (1 component), which in both cases was left temporal cortex. Four-component models were based on SVM method and involved sensorimotor cortex, left temporal cortex, posterior cingulate cortex/precuneus, and sylvian temporal cortex. Two-level discrimination as obtained by each model for each group is reported (within-group classification). Data are percentages.