Table 2.
Classification performance in ADNI held-out set and an external validation set.
| ADNI held-out (n = 90 individuals, 297 scans) |
NACC external validation (n = 1522 individuals, 2025 scans ) |
|||
|---|---|---|---|---|
| Deep learning model Area under ROC curve |
ROI-volume/thickness Area under ROC curve |
Deep learning model Area under ROC curve |
ROI-volume/thickness Area under ROC curve |
|
| Cognitively Normal (CN) |
87.59 (95% CI: 87.13–88.05) |
84.45 (95% CI: 84.19–84.71) |
85.12 (95% CI: 85.26–84.98) |
80.77 (95% CI: 80.55–80.99) |
| Mild Cognitive Impairment (MCI) |
62.59 (95% CI: 62.01–63.17) |
56.95 (95% CI: 56.27–57.63) |
62.45 (95% CI: 62.82–62.08) |
57.88 (95% CI: 57.53–58.23) |
| Alzheimer’s Disease Dementia (AD) |
89.21 (95% CI: 88.88–89.54) |
85.57 (95% CI: 85.16–85.98) |
89.21 (95% CI: 88.99–89.43) |
81.03 (95% CI: 80.84–81.21) |
Area under ROC curve for classification performance based on the deep learning model vs the ROI-volume/thickness model, for ADNI held-out set and NACC external validation set. Deep learning model outperforms ROI-volume/thickness-based model in all classes.