hi folks - I wanted to run by you an approach that I've been toying with
for doing functional connectivity (correlational) analyses in a random
effects setting. for each of 12 subjects I have four runs of a blocked
design with 2 conditions. what I would like to do is 1) find regions
that are correlated with a particular reference region, and 2) find
regions whose correlation changes significantly over the four runs. the
approach that I've been playing with is as follows. first, I run a
single fixed-effects analysis with all 12 subjects, and then use the VOI
function to extract the signal from the region of interest, correcting
for all effects of interest. then, I go through for each voxel in Y.mad
and calculate the correlation coefficient between the reference vector
and voxel timcourse, doing so separately for each run X subject. I put
these correlation coefficients back into image format, such that I have
48 correlation images (subject X run), and then run a random effects
analysis across the correlation images (either a one-sample t-test for
each run, or an anova across all four runs).
Does this sound like a reasonable approach? I'm particularly interested
in whether using the VOI data and correcting for effects of interest is
sufficient to deal with event-related transients, or whether there is
something that needs to be done to the voxel-by-voxel timecourses as
well. Also, if anyone has suggestions about better methods for this
type of analysis I'd love to hear them.
Cheers,
russ
--
Russell A. Poldrack, Ph. D.
Assistant Professor of Radiology, Harvard Medical School
MGH-NMR Center
Building 149, 13th St.
Charlestown, MA 02129
Phone: 617-726-4060
FAX: 617-726-7422
Email: [log in to unmask]
Web Page: http://www.nmr.mgh.harvard.edu/~poldrack
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|