Dynamical components analysis of fMRI data

We present a new multivariate analysis method for the analysis of fMRI data. This method tries to capture the deterministic structure present in the time series, using either an autoregressive scheme or the knowledge of the experimental paradigm, so that the interpretation of the spatiotemporal patterns is achieved in parallel with their detection. In the spatial domain, the components are made maximally independent through an ICA-like criterion. A global criterion is derived to express the model priors as well as the goodness of fit. The method is a priori adaptable to every sort of experimental conditions (block or event-related design). An experiment is presented on real data to show the potential of the method for the detection of signals, the analysis of their content as well as their localization.