Back to PLS HelpPosted on 09/22/17 09:02:50
Number of posts: 2
Dear everybody,
I am considering to use PLS to extract LVs that link common patterns of brain activity in different fMRI tasks and study whether such patterns differ across two groups of participants.
Every participants has three different session in which three different tasks are performed (I am only interested in the task condition of each task). Yet, two tasks are blockdesigns (8s on 10s off) and the other is an event-related design (2s epochs; 2-8 ITI + severak baseline periods x run).
Is it problematic to use PLS to study the LVs common to the three abovementioned tasks? In case it isn't: I) Should I model all tasks together - for each participant- and consider the different tasks as different runs of the same experiment? II) shoul I model everything as event-related PLS or as block-design PLS? III) In the event-related desig, is it problematic the overlapping of BOLD response between trials?
Thank you very much in advance and apologies if I have asked any nonsense
Didac
Replies:
Posted on 09/22/17 15:34:00
Number of posts: 394
Hi there - the idea is a good one, but it will be hard to do with the GUI version of PLS. Are you comfortable programming in MATLAB? The way to do it is do the analysis for the two experiments within each group and then correlate the results to determine the overlap. The correlation can be done with the singular images that come from the analysis and you can compute confidence intervals around the correlation to determine reliability. That would have to be done outside of PLSgui however. I can outline how to do that if you are able to code it.
Randy
Posted on 09/23/17 06:32:48
Number of posts: 2
Hi!
Thank you very much for the quick response.
Yes, programming should not be a (big) problem. Still, I would really appreciate if you can outline the main steps as I have no expertise in PLS analyses.
If I have correctly understood your answer; would you do:
i) independent analysis for each group*task (outside gui) in order to have a brain map for task*group condition.
Would you recommend any specific pipeline for processing each task and getting the group maps? I am especially concerned with the event-related task. Should I model it as a block design, or as an event-related (with HRF modeling?).
ii) introduce the brain values for each condition and introduce them in the PLS analysis (mean-centering task PLS?)
Thank you very much in advance,
Didac
Login to reply to this topic.
