Abstract

Principal and independent component analysis (PCA, ICA) are widely used in neuroimaging for explorative search for activation networks. I will review a statistical framework for PCA and ICA decompositions that allows us to perform model selection and evaluate components. I will discus some generalizations of PCA and ICA with significant potential in neuroimaging, including non-negative factorization, multi-way decompositions, and convolutive mixing.
Finally, I will introduce a set of Matlab based tools for explorative analysis of neuroimage data.