2021, Neumann, André, Peitek, Norman, Brechmann, André, Tabelow, Karsten, Dickhaus, Thorsten

We are considering the statistical analysis of functional magnetic resonance imaging (fMRI) data. As demonstrated in previous work, grouping voxels into regions (of interest) and carrying out a multiple test for signal detection on the basis of these regions typically leads to a higher sensitivity when compared with voxel-wise multiple testing approaches. In the case of a multi-subject study, we propose to define the regions for each subject separately based on their individual brain anatomy, represented, e.g., by so-called Aparc labels. The aggregation of the subject-specific evidence for the presence of signals in the different regions is then performed by means of a combination function for p-values. We apply the proposed methodology to real fMRI data and demonstrate that our approach can perform comparably to a two-stage approach for which two independent experiments are needed, one for defining the regions and one for actual signal detection.