Modeling of fMRI data

Abstract:
This work describes the use of statistical learning methodology in the
analysis of functional magnetic resonance imaging (fMRI) data.
Spatio-temporal support vector regression (SVR) is used in order to exploit
the intrinsic spatio-temporal autocorrelations in fMRI data.  The fMRI
signal arises from the nonlinear relationship between the blood oxygenation
level dependent (BOLD) signal and the underlying hemodynamics and neural
activity.  This framework is therefore inherently nonlinear and uses SVR to
extract the signal.  In our formulation we allow a compromise between a
prior model and the data.  We can examine the signal at multiple scales in
order to control the frequency content.  In addition, the formulation
allows the incorporation of multiple runs, subjects and tasks into the same
framework.