Clustering inter-subject correlation matrices in functional magnetic resonance imaging

We present a novel clustering method to probe inter-subject variability in functional magnetic resonance imaging (fMRI) data acquired in complex audiovisual stimulus environments, such as during watching movies. We calculate voxel-wise inter-subject correlation matrices across individual subject fMRI time-series and cluster them over the cerebral cortex. We address correlation matrix clustering problem and modify a standard K-means algorithm to cope better with spurious observations. We investigate suitability of the modified K-means with hierarchical clustering based postprocessing to correlation matrix clustering with several artificially generated data sets. We also present clustering of fMRI movie data. Preliminary results suggest that our methodology can be a valuable tool to investigate inter-subject variability in brain activity in different brain regions, such as prefrontal cortex.