Dear Seda,

Just as a side note, or in case this helps simplify your pipeline, you could simply select ROI-to-ROI when running gPPI analyses in CONN and select your choice of whole-brain parcellation atlas to produce and analyze whole-brain connectivity matrices for each task/condition directly within CONN.

Regarding task effects, it is perfectly fine to exclude task conditions during denoising if your planned connectivity analyses are only gPPI, as gPPI will always explicitly model task-effects as part of its General Linear model (independently of whether those gPPI analyses are run in or outside of CONN). For example, if you are planning to analyze in SPM the BOLD-responses for each task, and then use the peaks of these task-responses to define ROIs to then run ROI-to-ROI gPPI analyses with those peaks as ROIs, it is perfectly reasonable to exclude task-effects during denoising of your original functional data so that those effects are still present in the data when you use it for your SPM analyses. 

And regarding ROI extraction, my impression is that using spm_summarize for ROI extraction should be much more straightforward than using batch VOI tools (of course the latter allow you to define effects of interest and regress out effects of no-interest from the data, but it is not clear why you would want to do that if your data is already denoised)

Hope this helps

Alfonso

Originally posted by Seda Sacu:

Dear Dr. Nieto-Castanon and CONN experts,

I previously asked how I can find denoised ROI timeseries. After a more cautious look in the forum, I found that they are stored results/preprocessing. However, I have another related question. 

I am interested in task-based functional connectivity. My task has three explicit conditions. Since I am using whole-brain gPPI (doi: 10.1002/hbm.22532), I will conduct my first level analysis outside the CONN. As a second-level outcome, I have nxn matrix for each condition. That is, this is not a classic gPPI analysis (seed-to-voxel) but directed functional connectivity.

I have realized that effect of task is automatically included during denisoing step. I have seen a couple of posts that you are explaining the reasing behind this. It makes sense but I was thinking how this is compatible with SPM ROI extraction pipeline (https://en.wikibooks.org/wiki/SPM/Timese...). There, one need to adjust data for effect of task (using a F contrast), this tells SPM what is interesting in the GLM design (e.g., conditions) and regress out the other confounding variables (e.g., reliagnment parameters etc.). 

Actually, the method I use sounds similar to weighted ROI-to-ROI connectivity. If I am not wrong, you suggest regressing out the effect of task for this method too. If this is the case, I would also prefer regressing out the task effect. Nevertheless, I wonder whether if there is subtle difference between CONN and SPM in terms of modelling task effect for ROI extraction or if you tested this effect explicity.

Many thanks in advance!

Best,

Seda