PPI question stimulus onset times and TR grids

Hi there, I’m working through learning PPI analysis and have a question regarding step 2A on this document: https://afni.nimh.nih.gov/CD-CorrAna

2a. If your stimulus onset times were not synchronized with TR grids, pick up a sub_TR, e.g., 0.1 seconds, replace the above 1dtranspose step and upsample seed time series by xx (original TR divided by sub_TR) times:

My question is, how do I know if my stimulus onset times are synchronized with TR grids? Does this refer to whether the study was designed to have stimulus onset times starting and exactly aligning with the length of each TR?

If so, I definitely do no have osnet times aligned with my TRs.
My TR is 1.5 seconds, and I have a stimulus shown approximately every 4.5 seconds, and some variable resting periods within each run.

So the main question is how do I select a sub_TR? Is there some logic behind a sub_TR of .1, as given in the example?

If I were to select a sub_TR of .1, my timeseries would be upsampled by an integer of 15 (i.e. 1.5 TR / .1 sub-TR). Is that reasonable? I have no conceptual understanding of what this upsampling is doing, and how it solves the issue of not having synchronized TR grids and stimulus onsets.

I’d love any advice for selecting a sub_tr, and secondary, help understanding the conceptual basis for this!

Thanks a million

Following up on this question - still looking for clarification -thanks!


Okay, a year late is presumably better than never… sorry for letting this slip through.

Using 0.1 would be fine in this case. One effect of oversampling the time series is that it makes the deconvolution/reconvolution steps almost inverses, which avoids some side effects of the PPI filter. It also nicely allows for non-grid stimulus timing.

For reference, a slightly different example of a PPI analysis comes with the AFNI_data6 class data. The README.txt file is actually the main script. See AFNI_data6/FT_analysis/PPI.

  • rick