Perelman School of Medicine at the University of Pennsylvania

Section for Biomedical Image Analysis (SBIA)

participating with CBICA

MIDAS: Multivariate inference using discriminatively adaptive smoothing

Figure 1: Overview of MIDAS: 1) Local multivariate classifiers are applied to the brain, 2) Classifier weights are used to compute statistic, 3) P-value is obtained through analytical approximation


MIDAS (Multivariate inference using discriminatively adaptive smoothing) is a group analysis and regression framework that utilizes local classifiers or regressors, respectively, to obtain a test statistic for each voxel in neuroimaging studies. The significance of the test statistic is assessed through an analytic approximation of permutation testing (Figure 1).

MIDAS provides an increase in statistical power compared to state of the art methods by aggregating the information captured by local multivariate classifiers which act as local smoothers of signal(Figure 2).

Figure 2: AUC vs. sample size of MIDAS (in red) compared with other methods commonly used in neuroimaging group analysis



  • Christos Davatzikos
  • Erdem Varol
  • Aristeidis Sotiras


[1] Varol, Erdem, Aristeidis Sotiras, Christos Davatzikos. " MIDAS: Multivariate inference with discriminative adaptive smoothing." NeuroImage (In Preparation).