I am currently working on a project involving task-based fmri (3back tasks with only 3 back). I preprocessed the data using fmriprep and denoised the data using xcp_d. I used the custom-confound label in xcp_d to include Finite Impulse Response regressors, based on the paper:
Cole, M. W., Ito, T., Schultz, D., Mill, R., Chen, R., & Cocuzza, C. (2019). Task activations produce spurious but systematic inflation of task functional connectivity estimates. NeuroImage , 189 , 1–18. Redirecting
Since the 3back task contained correct and incorrect trials, I want to see if I can generate functional connectomes for correct and incorrect trials. I am currently thinking about the following two options:
- Clip and concatenate the correct trials together, accounting for ~5 seconds of hrf response, then generate the connectome; and do this for the incorrect trials as well.
- Pass subset FIR regressors into the design matrix in xcp_d to remove the undesired hrf for the condition. For example, to generate the connectome that focuses on the correct trials, my design matrix would include motion parameters that the denoising scheme requires, one set of FIR regressors to control for mean task-evoked activation, and lastly the FIR regressors generated with ONLY the incorrect onsets. And to generate the connectome that focuses on the incorrect trials, I would replace the FIR regressors generated with only incorrect onsets with the FIR regressors generated with only correct onsets.
The first option is what someone recommended me to do. And the second option is what I got from reading some papers on task-based fmri and thought might work. I am planning to use the first option as the last resort. But with the second option, I am not sure how the math will turn out, since the process will try to regress out some parameters twice.
If anyone has done this before, or knows how this will work, or have papers to recommend, or have better suggestions, please let me know!!! Thanks in advance!
P.S.: I thought about generating connectomes for 6 conditions+missed trials, accounting for congruent, incongruent, non-targets, but a lot of the subject had at least one of the condition missing, such as that they did not make any errors in one condition, or that they did not get anything correct in one condition. So I feel like it’s not that good, since my sample size is not particularly huge.