Figure 2.

Schematic showing the derivation of prior covariance components from fMRI data used in the PEB scheme. This entails the generation of a cortical surface mesh in participant's space [this can be obtained through the inverse normalization of a canonical mesh (see Fig. 1)]. The statistical map (SPM{.}) is thresholded to form an excursion set from which clusters are formed via connected component labeling. Each cluster is then independently projected onto the surface mesh using a 3D Voronoï diagram, leading to the definition of the same number of prior covariance components.