Print

Print
Chris
 
I would recommend using an effects of interest contrast to adjust the data and not a main effect of task. You can make the task choice when you form the interaction term with the PPI. Also the main effect of task that you choose for the interaction will be discounted when you set up your PPI design.
 
The effects of interest contrast would be
eye(12), i.e., ones down the main diagonal. I have left off the 13th column.
 
Again, while one could hypothesize self-connections as a source of variance this has nothing to do with PPI's. Furthermore if the eigenvariate is accounting for 95% of the variance in your VOI then it is unlikely that you have some large source of unmodeled variance.
 
What threshold are you choosing to select your VOI's?
 
Darren
 

 
On Thu, Mar 18, 2010 at 9:31 AM, Chris Watson <[log in to unmask]> wrote:
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