Assuming you have the three stage analysis (session/run, subject,
group), your first level (i.e., session/run) is the same as the
examples. The second level analysis (i.e., subject) is only using the
copes and varcopes estimated at the lower level. So, information about
your task timing and the timecourse of activity within a given region
is, at least as I understand it, no longer considered after the first
level (hence the "summary statistic" approach). I'm not sure how you
would go about doing a PPI without a physiological timecourse or an
event timecourse.
If you're worried about your first level not having enough time points
(how many do you have?), this would also apply to your other EVs, not
just the PPI. Theoretically, I suppose you could catenate the time
series from the runs from each subject, but this is not something I've
ever done or considered, so others would need comment on this idea...
Hope this helps.
David
On Wed, 2009-07-29 at 16:02 +0100, Regina Lapate wrote:
> Thanks for the reply- I am familiar with O'Reilly's page; however the first
> level in the case of my experiment (different from the examples given there)
> is the *session* and not the *subject*, hence my question below- regarding
> whether or not it would be valid to use such few time points to estimate
> correlations between brain areas; or whether it would be best to run the PPI
> at the second (subject) level instead (more time points, greater reliability).
>
> Thanks,
>
> Regina
>
> On Wed, 29 Jul 2009 01:37:27 -0400, David V. Smith <[log in to unmask]>
> wrote:
>
> >Hi,
> >
> >Take a look at Jill O'Reilly's page on PPI.
> >http://www.fmrib.ox.ac.uk/Members/joreilly/what-is-ppi
> >
> >In general, you'll need to model the PPI at the 1st level using the time
> >course of your seed region (derived from your pre-processed 4D data file)
> >and your task regressor.
> >
> >Cheers,
> >David
> >
> >
> >-----Original Message-----
> >From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
> >Of Regina
> >Sent: Wednesday, July 29, 2009 1:22 AM
> >To: [log in to unmask]
> >Subject: [FSL] PPI setup: session vs. subject level?
> >
> >Dear all,
> >
> >I am currently analyzing a study that I have modeled using the three-level
> >approach fsl recommends (session, subject, group) - where a session in this
> >case is simply a run of the same fMRI experiment- and now I would like to
> >run a PPI analyzes in it using a seed obtained from a group effect. My
> >question regards at what level one should set up the PPI in this case.
> >
> >It seems to me that, for PPI analyzes purposes, one should ideally run it at
> >the subject level (2nd), rather than at the session level (1st): given one
> >is interested in detecting correlations in activity between brain areas
> >dependent on task context, those correlations would tend to be less
> >reliable if they were estimated based on shorter time epochs (i.e.,
> >run/session) than if they were estimated based on a longer time epoch (i.e.,
> >the entire fMRI session), correct?
> >
> >If that is indeed the case (where the PPI setup should be done at the
> >subject level rather than at the session level, despite the fact the
> >experiment was originally modeled beginning at the session level)- which
> >'filtered_func_data' file should I use to extract the time course (using the
> >group effect seed) from? Currently, having adopted a fixed-effects model in
> >the intermediate level for each subject, I don't have a 'filtered_func_data'
> >for each subject as a result- and have instead one 'filtered_func_data' for
> >each cope of interest (feat directory) within each subject's .gfeat
> >directory. . . Would that (the 'filtered_func_data' from the cope of
> >interest for a given subject) be the one to use. . . ?
> >
> >Thanks much for your feedback,
> >
> >Regina
|