Hi Andrew & others,
As far as I know, the random effect analysis of fMRI data in SPM99
follows a two-level approach by first generating con_*.img's of a
contrast of interest within each session, and by then comparing these
images using the 'Basic Models' functions. For group comparison, it
seems to be important that only one image per subject-by-condition is
used. So when there are multiple runs of one condition, it is
suggested to create a single mean image and feed this into the
second-level analysis.
My problem is that there are motion artifacts in my data that require
me to discard individual runs (there are always 2 runs per
condition). Ergo, for some subjects, I am not able to create
mean-contrast images from 2 runs, but rather I am left with only 1
con_*.img. For the second level analysis I would then have to use
images that have a different 'history', (i.e.: some are averages from
2 runs and some are single con_*.img files).
I am afraid that this can be a major problem when doing a random
effects analysis, and in fact Andrew Holmes hinted at this 'caveat'
in previous mailings.
Another question pertains to the generation of the mean contrast
images: What is the best procedure to average two con_*.img? ? And is
there anything wrong with using ImCalc (evaluated function: (i1+i2)/2
for averaging two, already scaled, images)?
I would appreciate any comments and/or suggestions on these issues.
Thank you.
Thomas
--
---------------------------------------------------
Thomas Thesen
Department of Brain & Cognitive Sciences
Center for Visual Sciences
University of Rochester
Meliora Hall, Room 404
Rochester, NY 14627-0268
USA
phone: +1-716-275-0695
fax: +1-716-275 9216
e-mail: [log in to unmask]
---------------------------------------------------
|