Next: Bayesian framework
Up: tr04mw2
Previous: Overview
Model
Here we describe the GLM in the fully Bayesian framework. It will
be via the priors on the regression parameters that we propose to
constrain the possible linear combinations that are allowed.
Consider that the preprocessed FMRI data at voxel and at scan
is (
,
), the
row of the design matrix, , is , and
is a vector of parameter estimates. The preprocessed
FMRI data, , is taken to have been motion corrected and
high-pass filtered. The standard general linear model (GLM) is
then:
|
|
|
(1) |
We model the error, , as a voxel-wise
temporal autoregressive process
of order P (AR(P)). We can represent this as:
where is the AR coefficient (
).
Subsections
Next: Bayesian framework
Up: tr04mw2
Previous: Overview