Table 3.
Ability of the AD/HC models to predict MCI-to-AD conversion: morphometric data, ADNI cohort.
| Measurements | MCI-to-AD converters detection accuracya |
|||||
|---|---|---|---|---|---|---|
| 6 m (n = 14) | 12 m (n = 44) | 18 m (n = 30) | 24 m (n = 35) | 36 m+ (n = 26) | NC (n = 16) | |
| Cortical thickness | 78.6% | 79.5% | 70.0% | 62.9% | 65.3% | 75.0% |
| Sulcal depth | 71.4% | 77.3% | 53.3% | 51.4% | 53.8% | 62.5% |
| Jacobian | 60.2% | 70.5% | 76.7% | 71.4% | 53.8% | 56.3% |
| Volumes | 78.6% | 72.7% | 70.0% | 68.6% | 53.8% | 75.0% |
| Combined | 78.6% | 79.5% | 76.7% | 71.4% | 61.5% | 75.0% |
6, 12, 18, 24, 36 m+ — month of MCI-to-AD conversion; NC — non-converters (detected as HC by the classifiers); AD/HC — Alzheimer's disease/healthy controls.
a
Here the same models trained to classify images from AD and HC were used.