Identifying networks within an fMRI multivariate searchlight analysis

There is great interest in understanding how different brain regions represent information across space and time. Information-based searchlight analyses systematically examine the information encoded within clusters of functional magnetic resonance imaging (fMRI) voxels across the brain. Significant searchlights contain information that can be used to decode conditions of interest, but significant discriminability can be achieved in a variety of ways. We report a new analysis method that can identify sub-networks of searchlights based on having similar temporal changes in information. We present this method and apply it to fMRI data collected as participants viewed four visual categories: words, faces, shapes, and numbers. After running a searchlight analysis with a classifier, the resulting accuracy vector is submitted to a multi-subject Independent Component Analysis (ICA) that groups searchlights based on their decoding timeseries. The ICA identifies sub-networks of searchlights across variations in searchlight size, classifier, and binary versus continuous decoding metrics. These networks loaded onto visual and attentional networks, and onto predicted categorical regions of a probabilistic functional occipitotemporal atlas. In comparison, networks generated by representational similarity analysis were largely visually driven. These results demonstrate that this method can be used to divide searchlight maps into meaningful sub-networks.