In general detrending is just one of several denoising procedures that help compensate for very-slowly-varying trends in the data. High-pass filtering, and to a lesser degree also aCompCor and even regression of subject-motion timeseries, which are all part of the standard denoising procedure in CONN, can also contribute to remove slowly-varying temporal trends from the BOLD signal. So the general recommendation is to consider all these factors jointly, and if the BOLD signal after denoising (e.g. looking in the "GS after denoising" plot in CONN's denoising tab display, or in the associated carpet plots there as well) appears to still show the contribution of very-slowly-varying temporal trends, then consider applying a more conservative denoising procedure and choose the factor(s) that seem best matched to the observed residual effects (e.g. use higher-order polynomial detrending if the residual trends are well matched by low-order polynomials, use a higher-threshold in your high-pass filter if stationary low-frequency effects seem to better match the observed residuals, etc.)

And regarding your question about denoising for task designs, detrending (and all other denoising steps) is always performed separately within each individual session/run, and it applies to the whole timeseries within that run (e.g. it does not include trend-by-condition interactions).

Hope this helps
Alfonso
Originally posted by fmri questions: