Dear Dr. Gitelman and others,
The design is factorial. We have task/control X 2/3/4 choice (multiple
choice questions). 2x3 design.
When getting a VOI, it asks to adjust for effects. I chose "Main effect
of task", which would be the F contrast (temporal derivs included):
>> SPM.xCon(2).c'
ans =
1 0 1 0 1 0 -1 0 -1 0 -1
0 0
0 1 0 1 0 1 0 -1 0 -1 0
-1 0
[This would be 2task TD 3task TD 4task TD 2control TD .....]
I assumed this would be analagous to section 33.3 #5 in the SPM5 manual
(pp. 287). Then I chose a sphere of radius 4mm.
For the particular region I was asking about, it is in IFG/insula, which
I know is probably pretty heterogeneous; but I also see the same thing
when doing a PPI with BA6 (right in the center of FEF) as the seed
region, and also in the superior parietal lobe.
RE the variance: assuming the formula in line 283 of spm_regions is
correct, for a random subject:
>> xY.s(1)*100/sum(xY.s)
ans =
95.8826
(lowest of all subj's was 88.5%; the rest were 91.3% or higher)
So, would Hyoung-Ryul's ideas make sense? Would this be something to
explore in DCM (modulation of self-connections for those regions)?
Thanks,
Chris
Darren Gitelman wrote:
> Dear Chris
>
> I agree that is is not usual to see an activation at the site of the
> source region. PPI knows nothing about self or other connections per
> se so this doesn't explain the finding (although we could hypothesize
> various causes for the variance in that region) . I wonder if your
> eigenvariate is not a good representation of the activity in that
> region so that the interaction term (PPI.ppi) is picking up nearby
> voxels. When you extracted the VOI, how many voxels were included in
> it and how much variance did it represent? Also you say you adjusted
> for the main effects of task. Do you mean you included all task main
> effects (using an effects of interest type of contrast that only
> excluded nuisance effects) or that you adjusted the eigenvariate to
> only include variance from 1 particular task. The latter is generally
> not recommended (i.e., the eigenvariate should not be restricted to a
> single task as the source of variance but should include all real
> experimental effects).
>
> Darren Gitelman, MD
>
>
> 2010/3/17 강형률 <[log in to unmask] <mailto:[log in to unmask]>>
>
> Just to add an idea: maybe there is a task-related self-connection?
> In other words, the seed region's activity might have shown more
> quadratic autocorrelation during the task than during the control.
>
> PPI.Y can be different from the raw time courses of the BOLD
> signal of the voxels in the seed region because it is the first
> eigenvariate of them.
> Thus, the portion of the variance that correlated with PPI.ppi
> would have come from the rest of the variance, after excluding the
> first eigenvariate.
>
> Best regards,
> Hyoung-Ryul.
>
>
> On Thu, Mar 18, 2010 at 1:25 AM, Chris Watson
> <[log in to unmask]
> <mailto:[log in to unmask]>> wrote:
>
> Hello,
> In doing a group PPI, I see that in the results there is a
> significant cluster at the seed region I chose (in this
> example, BA9). Does this result make sense? [This is from an
> fMRI experiment; task vs. control]
>
> From this page
> (http://www.fmrib.ox.ac.uk/Members/joreilly/what-is-ppi) in
> Figure 1, it is stated:
> "Voxels in which activity is equally correlated with the seed
> region timecourse all the time will not show any correlation
> with the PPI regressor."
>
> Also in an email from Dr. Penny
> (https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind04&L=SPM&D=0&P=1605413
> <https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind04&L=SPM&D=0&P=1605413>):
> "if the beta value for I is significantly non-zero there is a
> psycho-physiological interaction ie. the psych variable
> changes the correlation between source and sink voxels."
> The psych variable shouldn't change the correlation between
> the seed region and itself.....
>
> I followed the steps in the manual (1. Get VOI in seed region,
> adjusted for main effect of task; 2. Do the PPI; 3. Specify +
> estimate design, w/ regressors PPI.ppi, PPI.Y, PPI.P). I would
> have thought that the PPI.Y regressor would explain the
> variance in the seed region...
>
> Any thoughts?
>
> Thanks,
> Chris
>
>
> --
> Hyoung-Ryul Kang, M.D.
> Seoul National University, Functional Brain Imaging Laboratory
>
>
