Dear Selma
> I am setting up a psycho-physiological interactions analysis and two things are not fully clear to me. Hopefully someone can help me out.
> My design has 3 conditions of interest for which I specified an F-contrast (memory load 1, 2 and 3) and a t-contrast for high load (3) versus low load (1). I want to extract the time series of an atlas-based ROI for which I have a mask. In the PPI analysis I want to compare conditions 1 and 3 (low and high load) in 2 atlas-based ROIs.
> 1) In the SPM tutorial and other examples I can find online, a specific contrast is selected for the voi time series extraction. What is the reason to select a contrast here? When creating the PPI variables you specify the conditions and contrast weights too. Therefore I think in my case, I would just want the time series adjusted for the F-contrast but without specifying a t-contrast. For now I just followed the SPM manual for RS DCM, so I added my ROI mask and the whole brain mask as mask.image and as expression i1&i2. I am not sure whether this is the right way to do it.
When extracting timeseries, two contrasts are typically provided to the software (they might be the same contrast). One selects the voxels to include (their timeseries are summarised using a PCA). The other tells SPM which regressors are interesting (the effects of interest F-contrast). Any non-interested effects are regressed out.
> 2) If I understand correctly, by adjusting for an F-contrast during voi time series extraction all other regressors are regressed out of the time series. In the SPM manual PPI example in step 36.5.1 PPI GLM analysis - Design setup and estimation, block regressors and a high pass filter are added to the model. Why would you need those regressors as they have already been regressed out during voi time series extraction?
Yes exactly - the effects of interest F-contrast cleans up the ROI timeseries for your seed region. Then in your PPI design matrix, you need those nuisance regressors again (high pass filter, head motion etc), because those effects probably contribute to the target voxels that you are modelling all over the brain.
Let us know if anything remains unclear.
Best
Peter
|