Table 2.
Imaging response detection performance of the null models
| Predictor set | Support vector machines | Extremely randomised trees |
|---|---|---|
| Brain volume | 0.469 [0.463–0.475] | 0.498 [0.490–0.506] |
| Number of lesions | 0.457 [0.449–0.465] | 0.464 [0.458–0.470] |
| Total lesion volume | 0.473 [0.465–0.480] | 0.535 [0.528–0.542] |
| Best low dimensional | 0.497 [0.490–0.503] | 0.473 [0.467–0.480] |
| Regional atrophy | 0.445 [0.437–0.453] | 0.454 [0.447–0.461] |
| Regional disconnection | 0.465 [0.457–0.473] | 0.467 [0.459–0.476] |
| Best high dimensional | 0.459 [0.450–0.469] | 0.422 [0.414–0.430] |
None of the null models could reliably detect the intervention. All figures are given as mean AUC [95% CI]