On Mon, Jul 16, 2012 at 1:07 PM, Sophie Forster <[log in to unmask]> wrote: > Dear SPMers, > > I am working on a gPPI analysis but am stuck on a few specific issues > about adapting the example wrapper and input structure. I already > searched the SPMlist and couldn't find the answers to these questions > so I wondered if someone out there might be able to advise me? I can help with your questions. > > To give the background, my original (event-related) model has 10 event > types, I ultimately want to look at the PPI for at least two contrasts > from this model, and using seed regions based on the mean across a > predefined ROI. I am ultimately looking at individual differences so > it is particularly important that the VOIs for each subject are > equivalent (this was a problem I ran into when trying to make the VOIs > using the 'common method' of VOI extraction as discussed in the email > exchange forwarded below). From what I read on the SPM list, gPPI > seems to be a more suitable than the standard PPI toolbox in SPM for > achieving these aims. There are several differences between the standard PPI method implemented in SPM and the that implemented in the gPPI toolbox. I should note that the standard SPM approach can be run through the toolbox. Also, the gPPI approach can use individualized VOIs based on activity with a standard VOI (although I prefer the mean from the entire VOI for consistency). The main difference is in the creation of the regressor and is detailed in the NeuroImage paper (McLaren et al. 2012). I have been following the instructions > regarding editing the ppi_wrapper and templatemaster from here: > > http://brainmap.wisc.edu/PPI These instructions haven't been updated in awhile, but will still work. I'm currently working on the manual. > > My questions are: > > 1. Am I right in understanding that P.Tasks in the template_master > file refers to the event types in my model? If so, should I list only > the events that I am interested in looking at the contrasts for, or > all events? For example, if I'm only interested in the third and fifth > event in the model, called 'event3' and event5', would I set P.Tasks = > {1 'event3' 'event5'} I generally include all tasks in P.Tasks. The one exception would be when you have very few trials of one condition, then you might want to leave it out. The effects of modeling some conditions versus all conditions has not been fully studied. P.Tasks uses that task names from the first level. The names are those that are found in SPM.Sess.u.name. > > 2. P.Weights: I am assuming if I have listed only two events in P. > Tasks I only need to weight these? So, if for example P.Tasks is set > as above, and I want to look at a contrast of event3-event5, would > P.Weights = [1 -1]? Or should I not be weighting this? If you are using the 'cond' option - which is gPPI, then there is no need for P.Weights. P.Weights is only needed if you want to use the PPI approach from SPM where the subtraction is done before estimating the model. Given the complexity of your design, I would discourage the use of the standard PPI approach. Set P.method to 'cond'. > > 3. When I try to run the ppi_wrapper script, with the template set as > above, it seems to get stuck on P. Contrasts, with error messages > saying that P.Contrasts.left and P.Contrasts.right are not formatted > correctly, and that P.Contrasts cannot be defined automatically. I am > unsure as to what I should be entering into this variable? P.CompContrast was set to 1; however, you did not define P.Contrasts properly. Since you have many tasks, P.Contrasts cannot be defined automatically. P.Contrasts is the contrasts that you want to generate after creating the model. These aren't too hard to create after you know the structure. (1) Determine the null hypothesis: cond1=cond2 (2) P.Contrasts.left={'cond1'} %cond1 is the name of the actual condition on the left side of the equal sign (3) P.Contrasts.Right={'cond2'} %cond2 is the name of the actual condition on the right side of the equal sign (4) P.Contrasts.MinEvents=5 %This is the minimum number of events needed. I generally try to aim for 30 events per condition (5) P.Contrasts.Name='name_of_contrast' (6) P.Contrasts.Weighted=5; %determines if trials will be weighted by the number of events in each run or by the number of runs. Anything condition with an event longer than 5 seconds will lead to weighting by the number of runs. If all events are shorter than 5 seconds, then the runs will be weighted by the number of events in each run. The other fields can be left blank. If you want more than one contrast, then repeat the above, but use P.Contrasts(2), etc. for more contrasts. > > 4. Am I right in assuming that, if I enter a .nii file as my ROI, this > mask will be used to create individual VOI time-series for each > subject which will then be used as the seed region for the PPI? I > couldn't find where this initial part happens in the script so just > wanted to check I am understanding the process correctly. For each subject, the mean time-series will be extracted from this ROI. It is important to note that the ROI will be the same for each subject. > > 5. More minor point but I am a little confused about how the wrapper > loops through seed regions. If anyone has an example of the > ppi_wrapper script that they edited and got working that they wouldn't > mind forwarding to me I would find that very helpful. If you look at the example wrapper, you will see that there is a loop over the number of subjects. ppi_wrapper(region,firstsubject,lastsubject) can be run across 35 subjects in the first ROI by using: ppi_wrapper(1,1,35) Just define Subjects, RegionFile (nii file), and Region name. > > Any advice would be greatly appreciated! > > Many thanks and best wishes, > > Sophie Let me know if you have any other questions. > > > > ---------- Forwarded message ---------- > From: MCLAREN, Donald <[log in to unmask]> > Date: 23 February 2012 15:09 > Subject: Re: [SPM] VOI extraction for PPI analysis > To: [log in to unmask] > > > There are a number of ways to define VOIs. I've commented on your > methods and described my own below. > > Best Regards, Donald McLaren > ================= > D.G. McLaren, Ph.D. > Postdoctoral Research Fellow, GRECC, Bedford VA > Research Fellow, Department of Neurology, Massachusetts General Hospital and > Harvard Medical School > Website: http://www.martinos.org/~mclaren > Office: (773) 406-2464 > ===================== > This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED > HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is > intended only for the use of the individual or entity named above. If the > reader of the e-mail is not the intended recipient or the employee or agent > responsible for delivering it to the intended recipient, you are hereby > notified that you are in possession of confidential and privileged > information. Any unauthorized use, disclosure, copying or the taking of any > action in reliance on the contents of this information is strictly > prohibited and may be unlawful. If you have received this e-mail > unintentionally, please immediately notify the sender via telephone at (773) > 406-2464 or email. > > > > On Thu, Feb 23, 2012 at 9:25 AM, Matthias Schmidt <[log in to unmask]> wrote: >> >> Dear colleagues, >> >> I have two questions about VOI extraction from a source region for PPI analyses: >> >> 1) A common procedure is to identify a peak voxel of a contrast of interest and define a sphere around this voxel (chapter 33.3 and 33.4 ). > > > The sphere is drawn based on the single subject peak and only includes > significant voxels. > >> >> However, I have a second approach, which I want to discuss with you: >> >> E.g. you are interest in a PPI analyses with the seed voxel in the amygdala. >> First, you create an amygdala mask for each subject individually containing only significant voxels in the amygdala (derived from a contrast of interest). After that, you extract the VOI data of these voxels and not of an arbitrary sphere. What do you think of this approach? > > > I like the approach of using a mask and looking at things inside the > mask; however, as you point out in #2, finding significant areas might > not yield overlapping areas. The larger the mask, the more likely you > could get different areas. In your case, you might have some subjects > with activity in the ventral amygdala and some with activity in the > dorsal amygdala. If you have two groups and the groups are split by > the location of activity, then you could bias the connectivity > results. I generally just use a single mask and grab all voxels in the > mask for all subjects. I also compute the mean, rather than the > eigenvariate as not to skew the contributions of different voxels in > different subjects. > >> >> 2)If you use the ‘common’ procedure for extracting the VOIs (define a sphere around a voxel), you define a p-value adjustment (e.g. FWE .05) as can bee seen in chapter 33.4 (Point 1). However, this voxel is below this threshold in some subjects and if you press “eigenvariate” SPM will take the next neighboured voxel, reaching the threshold. However, sometimes this voxel is really far from the originally designated voxel. How do you handle this problem (e.g. decrease the threshold)? > > > This is a big problem, with no easy solution -- see comments above. > > The gPPI toolbox allows various region definitions including the ones > that you have mentioned here (http://www.nitrc.org/projects/gppi). > > >> >> >> Best, >> Mat >> >> >> -- >> Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir >> belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de