Table 1.
Imaging response detection performance of each model and the two types of classifier
| Predictor set | Support vector machines | Extremely randomised trees |
|---|---|---|
| Brain volume | 0.554 [0.545–0.563] | 0.595 [0.588–0.602] |
| Number of lesions | 0.550 [0.540–0.561] | 0.601 [0.594–0.609] |
| Total lesion volume | 0.626 [0.618–0.634] | 0.635 [0.629–0.641] |
| Best low dimensional | 0.647 [0.639–0.655] | 0.686 [0.679–0.693] |
| Regional atrophy | 0.857 [0.852–0.862] | 0.819 [0.813–0.825] |
| Regional disconnection | 0.817 [0.810–0.824] | 0.822 [0.815–0.828] |
| Best high dimensional | 0.869 [0.864–0.873] | 0.890 [0.885–0.895] |
The best high-dimensional model was constructed as one which provided an average of the predictions made by the regional atrophy and the regional disconnection models, weighted by their corresponding mean AUCs. All figures are given as mean AUC [95% CI]