Clustering of spatiotemporal signals: application to the analysis of FMRI data

In this paper, a new paradigm for the analysis of event-related functional magnetic resonance images (fMRI) is explored. These datasets are large collections of spatiotemporal time series that are indexed by their spatial locations. The analysis of such datasets requires to partition the spatial domain into regions within which the time series are similar. Basis functions is constructed on which the projection of the fMRI time series reveals the organization of the fMRI dataset into "activated", and "non-activated" clusters. These basis functions maximally separate the time series into meaningful groups according to their time-scale behavior.