In this study, we aimed to reveal networks of brain regions that were functionally connected during object-related audio-visual (AV) integration. To this end, we used spatial independent component analysis (sICA), a multivariate, data-driven analysis technique that decomposes an fMRI dataset into spatially independent components, which can be interpreted as functional connectivity maps. Each component was associated with a single time course. These component time courses were then tested using the knowledge about the stimulation time course, thus enabling a classification of components as mainly visual, auditory, or multisensory, amongst others (like physiological components related to breathing and heart beat, or otherwise unmodeled sources). Regions-of-interest (ROIs) were defined as clusters of voxels which contributed significantly to at least two of the three above-mentioned components of interest, in other words, they were the regions of overlap of the component maps.
Voxel time courses can be thought of as the sum of all component time courses weighted by the values of the respective component maps at that voxel. This can result in a variety of voxel characteristics: Voxels in a region where only one spatial component has large map values will show a time course very similar to the respective component time course, e.g. mainly auditory or visual activity. Due to the weighted mixing of components, both visual and auditory unisensory components can contribute equally to a voxel time course. In a GLM analysis the effects of auditory and visual stimulation may be simply additive at this voxel. If the mixing comprises non-zero coefficients for components that describe purely multisensory processing, i.e. processing that is absent during unimodal stimulation, the respective voxel will show superadditive effects. Thus sICA provided us with a set of hypotheses regarding these ROIs.
In order to explicitly test these hypotheses we analyzed the data of an independent experiment 2 in these ROIs using the knowledge of the stimulation time course of this experiment via a massively univariate (hypothesis-testing) voxel-based GLM. Applying the max-criterion (i.e., 0<A<AV>V>0) revealed robust AV integration effects in the ICA-defined bilateral cortical network (consisting of pSTS, VOT, PPC, and PFC regions), thus demonstrating the sensitivity and value of the proposed analysis approach.