Hey AFNI folks,
I'm running a resting-state functional connectivity study and am trying to follow the methods outlined in a paper by Power et al. 2014 (doi:10.1016/j.neuroimage.2013.08.048). As a result, I am wanting to do bandpass filtering after regression of nuisance variables (motion, csf, etc.) has been carried out.
I do the regression using 3dDeconvolve and use the -censor flag to designate what volumes to use/ignore in the regression. Although no sub-bricks are actually removed, the output of the -errts residual file has all zero voxels for sub-bricks that were censored. All is good (I think) so far...
Next step, I run 3dBandpass to do the filtering. After this, every sub-brick (including the ones that were previously all zeros due to censoring) are filled in, with what at least appears to be, reasonably looking data. My question is: is this a trustworthy way of running the bandpass filtering? Do those all-zero sub-bricks totally screw up the filtering or is AFNI doing something smart behind the scenes to deal with this? Finally, if this is not a kosher way of doing the filtering, how would you recommend moving forward in terms of filtering the data? Should I use the 1d_tool.py to extract only the non-censored timepoints and do the filtering only on those sub-bricks? This also doesn't seem right...
Sorry for the rant! Any help would be much appreciated!!!