 |
 |
On Thu, Jan 5, 2012 at 7:36 AM, Ben Becker <[log in to unmask]> wrote: > Dear SPMer, > > I want to analyze differences in connectivity in two groups of patients & > got some basic questions regarding the design: > My fMRI design: > > - event related > - participants see emotional & neutral pictures > - after scanning participants had to remember the images > - first level GLM matrix has the following regressors: > > 1.emotional remembered (ER) > 2. emotional not remembered (EnR) > 3. neutral remembered (NR) > 4. neutral not remembered (NnR) > > - second level group comparison reveals higher activity of the > patients for the contrast (ER > EnR) in the insula > Now I want to explore differences in connectivity for ER > EnR between the > groups using the seed region insula > > My PPI questions: > > 1. Extracting VOIs > > a) Which contrast should be used for extracting VOIs? > Remembered > not remembered? > > You can use whatever contrast you want to define your regions. PPI is not specific to a contrast, but rather a region. The choice of region is up to you. I usually just draw a sphere around the peak coordinate. Other use the entire cluster and some threshold the region for each individual subject based on the first-level maps. Each of these has advantages and disadvantages that have been discussed previously on the list. > b) During extracting VOIs: should data be adjusted for > effects for interest > (F:contrast > 1 0 0 0 > 0 1 0 0 > 0 0 1 0 > 0 0 0 1) > This is the correct contrast to adjust for. > > 2. Create PPI variable > > Include ER -> yes 1 > Include EnR -> yes -1 > Include NR -> no > Include NnR –> no > I would highly recommend that you use gPPI and include all four tasks in the model. It is clear from simulations that you can induce false positives when you only model 2 conditions as a subtraction when you have 4 conditions. In particular. If you assume that ER-EnR=0 and NR=.5, add random noise to mimic multiple subjects, you can find that ER is different than EnR. These findings are included in my paper that is under review (submitting the revisions soon). gPPI is also automated (extraction, creating the model, estimating the model, and creating contrasts all in one step) and allows all the options you get in SPM (variety of region definition options and the use of the subtraction method, despite its potential problems with false positives. gPPI is available at http://www.martinos.org/~mclaren/ftp/Utilities_DGM > > > Thanks in advance & happy new year to all! > > Ben >