Hi Katie,
Don't model fixation (it's an implicit baseline). You shouldn't have a case where every time point of your experiment is modeled.
You should turn on temporal filtering for any task EV. Assuming you've done temporal filtering on your pre-processed data, you want to apply the same filter to the model (except for the phys regressor, which was presumably derived from pre-processed data that was temporally filtered).
I don't think you should include an additional regressor to account for the global signal because it would be heavily correlated with your phys regressor. However, I am curious what others think about the idea of including the global signal and orthogonalizing the signal from a particular seed region w.r.t. that global signal before setting up the PPI regressor. Although there's been some debate about this issue in the past year or so, several folks who do resting-state connectivity analyses generally take a seed region (e.g., PCC) and force that signal to be orthogonal w.r.t. the whole brain signal. It seems this basic approach could be extended to task-based connectivity analyses like PPI, but I suspect there has to be a reason for not doing this...
Cheers,
David
On May 4, 2010, at 11:45 AM, Kathryn Cullen wrote:
> Dear David,
> Thanks so much for your help. I was able to get the correct version to then apply the correct "zeroing" options.
>
> Would you mind telling me if my PPI set up is coorect: For EV1, I have the task, set up as a 3 column entry, with 0 for fixation (baseline), 1 for fear and -1 for neutral, to hopefully allow me to look at fear minus neutral in this EV. For EV2, I have the timecourse. For EV3, I have the interaction. I have no filtering on any of these EVs, but I do have the task EV convolved with the "gamma" otpion. I have no additional EV regressors (not sure what these would be for my study design), and for the contrasts I have:
>
> EV1 EV2 EV3
> 1 0 0
> 0 1 0
> 0 0 1
>
> Should I be adding an additional regressor to take out the global signal? Anything else I'm doing wrong here?
> Thanks very much for your guidance in my probably very rudimentary questions.
>
> -Katie
>
> On May 3 2010, David V. Smith wrote:
>
>> Hi Kathryn,
>>
>> 1) Which version of FSL are you using? You should have three 'zeroing' options (min, centre, mean). They are not the same (see http://www.fmrib.ox.ac.uk/Members/joreilly/the-zero-ing-options-in-ppi). If your GUI display doesn't look like the ones on Jill O'Reilley's PPI pages, then you may need to grab the latest version of FSL.
>>
>> 2) Assuming you're just contrasting the face conditions and leaving the fixation unmodeled (it's your baseline, right?), then you should be able to combine the two regressors into one by putting 1s and -1s in the 3rd column of the regressor. See http://www.fmrib.ox.ac.uk/Members/joreilly/frequently-asked-questions-ppi for more information.
>>
>> Hope this helps,
>>
>> David
>>
>>
>>
>> On May 3, 2010, at 6:52 PM, Kathryn Cullen wrote:
>>
>>> Hello, I am trying out a PPI analysis using FEAT for the first time and have run into 2 questions:
>>> (1) In the instructions that I have seen, there is an option for "make zero" on the stats set up where you choose "interaction" for the 3rd EV, but in my fsl gui there is only the option to check "demean". Is it OK to check demean for both EV1 and EV2, and will it mean the same as what's described in the on-line presentation (Centre for EV1 and mean for EV2).
>>> (2) The task for this experiment involves 3 conditions, which are viewing fixation, viewing neutral faces, and viewing fear faces. I'm not sure how to set up the contrasts in FEAT since the examples are usually for on-off paradigms. This is a problem that I also am running into as I try to run the traditional analysis of activation between conditions and between patient groups.
>>> Thanks, Katie
>>> Kathryn R. Cullen, M.D.
>>> Assistant Professor
>>> University of Minnesota Medical School
>>> 2450 Riverside Avenue Minneapolis, MN 55454
>>
|