Optimal statistical analysis of functional magnetic resonance data

Project members: J. Sijbers (University of Antwerp), A.J. den Dekker, A.M. Van Der Linden (University of Antwerp), D. Van Dyck (University of Antwerp)

Sponsored by: FWO

Functional Magnetic Resonance Imaging (fMRI) is a relatively new technique for functional imaging of the brains, or the detection of active regions within brains under a variety of experimental setups. By means of a series of functional MR recordings during which a task model is applied, the aim is to find the hemodynamic response in order to constitute a statistical parameter map (SPM) of significantly active regions for each experimental setup.

Nearly all signal processing methods currently applied to fMRI data, such as construction of an SPM, are developed for gaussian distributed data. fMRI images, however, are not gaussian but Rician distributed. An important goal of our research is the development of statistical tests that account for the true, underlying data distribution. Previous research has already demonstrated that estimation of parameters from MR images may be improved significantly in case the incorrect assumption of gaussian distributed data is no longer retained but the correct (Rice) distribution of the data is taken into account. Therefore, it is expected that also this research will prove its value for the computation of SPMs.

Furthermore, a new technique will be developed that exploit spatial correlation of functional active regions within the image.