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 > >