An introduction to Multiway Methods for Multi-Subject fMRI experiment.
FMRIB Technical Report
Didier G. Leibovici and Christian Beckmann
Oxford Centre for Functional Magnetic Resonance Imaging of the Brain (FMRIB),
Department of Clinical Neurology, University of Oxford, John Radcliffe Hospital,
Headley Way, Headington, Oxford, UK
[didier]or[beckamnn]@fmrib.ox.ac.uk
Model free exploratory methods available for functional MRI analysis of one subject such as
Principal Component Analysis (PCA) and Independent Component Analysis (ICA) are investigated for
generalisations when analysing more than one subject. Introducing the subject dimension makes a
three-way data: brain, time, subject, and multiway methods are proposed either to optimise
the variance (PCA) or the Negentropy (ICA). Results of folded methods and purely multiway ones are
described on a 12 subjects verbal study.