Mining Interaction Patterns among Brain Regions by Clustering

Functional magnetic resonance imaging (fMRI) provides the potential to study brain function in a non-invasive way. Massive in volume and complex in terms of the information content, fMRI data requires effective, and efficient data mining techniques. Recent results from neuroscience suggest a modular organization of the brain. To understand the complex interaction patterns among brain regions we propose a novel clustering technique. We model each subject as multivariate time series, where the single dimensions represent the fMRI signal at different anatomical regions. In contrast to previous approaches, we base our cluster notion on the interactions between the univariate time series within a data object. Our objective is to assign objects exhibiting a similar intrinsic interaction pattern to a common cluster. To formalize this idea, we define a cluster by a set of mathematical models describing the cluster-specific interaction patterns. Based on this novel cluster notion, we propose interaction K-means (IKM), an efficient algorithm for partitioning clustering. An extensive experimental evaluation on benchmark data demonstrates the effectiveness and efficiency of our approach. The results on two real fMRI studies demonstrate the potential of IKM to contribute to a better understanding of normal brain function and the alternations characteristic for psychiatric disorders.