Modelling Strategies for Functional Magnetic Resonance Imaging

This thesis collects research done on several models for the analysis of functional magnetic resonance neuroimaging (fMRI) data. Several extensions for unsupervised factor analysis type decompositions including explicit delay modelling as well as handling of spatial and temporal smoothness and generalisations to higher order arrays are considered. Additionally, an application of the natural conjugate prior for supervised learning in the general linear model to efficiently incorporate prior information for supervised analysis is presented. Further extensions include methods to model nuisance effects in fMIR data thereby suppressing noise for both supervised and unsupervised analysis techniques.