Fellow Travelers,
We're currently considering new analysis paths for task-based FMRI data and would like to hear your thoughts with respect to applying to task data the same denoising procedures that we might apply to resting-state data. As I understand it, the potential disadvantage of applying, say, the same denoising steps that are carried out in afni_proc example 5.c. would be loss of degrees of freedom. The advantage would be reducing noise and improving the sensitivity of our analysis. A few questions arise:
1. Does losing degrees of freedom *really* compromise the sensitivity of most analyses? At an intra-subject level this is the case, but in most analyses (at least among the folks I hang out with), voxel-wise fit coefficients (and sometimes error estimates--e.g., 3DMEMA) are normalized to standard space and combined in a group-level analysis and there's no "penalty" that carries with them (nor should there be). Yeah?
2. If applied properly, there's little chance in the context of intra-subject task-based analysis of typical RS-FMRI noise covariates *incorrectly* encroaching on task-based variability in FMRI time-series, thereby decreasing the sensitivity of our analysis. Yeah? Stimulus-correlated motion and respiration would count as noise covariates that should encroach on task covariates but, if those factors aren't in play, then the more noise we take out, the better. Truesies?
3. The devil of this is likely in the data, themselves. Have you or do you know of a group that has examined effects on sensitivity of task-based designs of applying varying levels of denoising up to and including what we currently apply to RS-FMRI data? Also, I have to imagine this discussion has come up on this forum (although my search through the message board didn't turn up much) so please feel free to direct me elsewhere.
Thanks!
Paul