Hello, I'm currently using an AFNI pipeline to model some multi-echo task data, and it's going great! However, I wanted to model RT, and I'm a tad confused. I'm very new to the gritty details of modeling the fMRI signal and AFNI in general, but I've seen a couple papers (Grinband et al., 2008; Mumford et al., 2024) that mention an effective way to model RT is to include a regressor for the RT of each event with a duration equal to the RT of each response. I understand how to do this is theory given the presence of several sustained, married regressors in the model already. My task contains 3 general phases: a lead up section of basic trials, a subtask phase that make up the block of interest for my subsequent contrasts, and a sort of cool down phase of more basic trials.
My questions are:
1- Would I include a single regressor that encompasses every single trial across all 3 phases or only model subtask trials (the second phase of the described task)? I worry that limiting the regressor only to "trials of interest" within my subtask phase may introduce some overlap and subsequent correlation between the new RT regressor by condition and the sustained regressor attempting the capture the neuronal activity of each condition.
2- Assuming my intuition is correct relating to question 1, would this just require a single new text file (ex. RT.txt) that includes a row of onsets:durations for each trial for each run? I have 6 runs, so the final product should be 6 rows of extremely long pairings of onsets:durations correct?
3- Is there anything else I should know as someone who is relatively new to all this relating to modeling RT? I know there are other methods described in the papers above, but I believe the method I've described thus far is the most apt. However, I'm happy to be proven wrong if need be.
To clarify, are the three phases contiguous? Are you planning to estimate the BOLD response for each phase individually?
Incorporating behavioral data, such as reaction time, is a complex issue. Its importance is debatable and much nuanced than what is presented in the literature. If your research hypothesis does not focus on the association between BOLD response and behavior, I recommend initially modeling the data without considering the behavioral data and using that result as your baseline.
Assuming my intuition is correct relating to question 1, would this just require a single new text file (ex. RT.txt) that includes a row of onsets:durations for each trial for each run? I have 6 runs, so the final product should be 6 rows of extremely long pairings of onsets:durations correct?
That seems fine. Technically, if reaction time is only relevant to the second phase, you only need to provide the RT information for the stimulus timing of that phase.
If you don't mind, I would like to hear your impressions on how the results compare between models that incorporate the RT data and those that do not.
Thanks for the response! Yes, the phases are contiguous. The boundaries between them are cue trials more or less but are still in sequence. I have modeled the data without it. However, I have 2 conditions that show marked behavioral differences that I'd like to make a claim differ by the processes of interest. I believe the behavioral differences may obfuscate which voxels are related to the different processes, and I do see the more behaviorally challenging condition activating a much wider, general swathe of brain areas.
Also, while the 2nd phase is the only area of interest for later analyses, I model each part of the task to account for the most variance as possible.
I have 2 conditions that show marked behavioral differences that I'd like to make a claim differ by the processes of interest. I believe the behavioral differences may obfuscate which voxels are related to the different processes, and I do see the more behaviorally challenging condition activating a much wider, general swathe of brain areas.
The typical implementation of modulation by a behavioral measure aims to account for cross-trial variability within each condition. If your goal is to also account for variability across conditions, this may involve a mediation process, resulting in a complex situation that typical modulation is not designed to address.
Hmm. I'll look more into this. I successfully seem to have ran through a subject with Rt as a regressor. Although now I have a rather humorous new problem in that the number of sub-bricks that appear when I select the Olay or threshold in the afni viewer is too high to see all of them on my screen thus making it impossible to choose the last ones that are the contrasts of interest. I found a 20 year old post suggesting there should be a scrollable pop-up window to select which sub-brick I'd like, but I can't seem to find it. Any suggestions?
Edit: I found it as soon as I finally hit post of course. Nevermind! Thank you.
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.