“fMRIPrep does high-pass filtering before running anatomical or temporal CompCor. Therefore, when using CompCor regressors, the corresponding cosine_XX regressors should also be included in the design matrix.”
Does it mean CompCor regressors should not be use independent of cosine_xx regressors? What happens if you did include that? Also is cosine_00 correponding to a_comp_cor_00 and so on? It seems that the numbers don’t match (e.g. I only had cosine_00 and cosine_01) but up to a_comp_cor_14 in one session.
If you do not include cosine regressors, your GLM will not account for low-frequency drift over the course of the scan. If you include a drift term that has a different filter cut off, then you might not account for all the low frequency drift in the data.
No, these are independent terms. The number of cosine regressors is based on your TR. The number of compcor components is partly determined by how long your time series is, but since it is a stochastic data driven approach, the number of components may vary by run.
Thank you for your prompt reply. what is the effect of not including the low-frequency drift and what is your filter cut off? Does it just make the data noiser or was it creating a systematic bias?
No, these are independent terms. The number of cosine regressors is based on your TR. The number of compcor components is partly determined by how long your time series is, but since it is a stochastic data driven approach, the number of components may vary by run.
What does it mean by “the corresponding cosine_XX regressors” then? Should we include all cosine_XX regressors if using any compcor components?
Scanner heat usually leads to systematically higher signal intensity over the course of the scan. You want your signal to be most related to the task at hand, and not biased by when in the scan the task occured.
128 seconds for acompcor.
Corresponding is perhaps not the best word, as there’s no 1-to-1 correspondance. But these are the same regressors used to filter data before acompcor.