Parametric Modulation with fMRIPREP- odd #0 contrast results

AFNI Message Board
1

I am trying to do a parametric modulation with 3 different modulators on the same event. I am using AM2 and the data was preprocessed using fMRIPrep so most of the steps are completed before afni proc. When i review my results the contrast with the #0 looks strange ,as in a full covered picture of activation and no clusters. The other contrasts look better with the #1, but it also seems like not all of the different modulators are showing up in the results. I would be happy to send over the stats file as well as the setup scripts and 1D image if someone would be willing to check that I did not mess something up in this process. I cannot upload anything this large in the file attachment section on the help board.

2

Hi AFNIi Help,

Just wanted to follow again on this as i have not heard from anyone yet from my original post on 7/11. I would be happy to set up a zoom meeting if that is more useful or send over things if someone could share a file upload that is larger than the forum allows. This is rather URGENT as I have dissertation and conference deadlines relying on this data getting processed correctly! Again the issues is viewing the results from the #0 contrast showing just full square of activation over the entire brain and outside, vs the #1 contrast showing the usual activation clusters.

Thank you,
Tessa

3

Hi Tessa,

Please feel free to create a tgz package with that information, including the *.xmat.1D file, and send that to me.
Click on my name for the address.

  • rick
4

Hi Tessa,

After looking at this in some detail, while still a bit confusing, it seems that the apparent full-volume stats are coming from inclusion of the compcor regressors. If those terms are not included in the model, the results look more as expected (though I do not know exactly what to expect from your analysis).

Getting the proper AM stim types and such would be good, but I do not think that really affected anything (except that the requested contrasts are surely not what you want).

Anyway, I think the strange results have something to do with a global signal type of regression (where the compcor components have task signal included).
But it is still very strange to get such a strong result across the entire volume. I can imagine how that might happen, but it seems so unlikely…

Try re-running your analysis without them included, and see how it looks.

  • rick
5

Hi Tessa,

Just as a follow-up to others reading this thread, Tessa and I have looked into this and found that there is a bug in 3dDeconvolve when using -stim_times_IM for a regressor class that has no events. It is not such an odd case - one can see the use of IM in the case of missing or incorrect reponses, say, and then imagine that for some subjects, there might not be any such events. Having no events is fine for other cases, like -stim_times or -stim_times_AM2, but there are problems with -stim_times_IM.

We will look at what might be going on here. More news at some point…

Thanks a lot, Tessa!

  • rick