Dear AFNI minds,
3dDeconvolve can accomodate (model) low-frequency signal drifts using -polort. In my case, I have a task that is about three 7-minute runs concatentated. AFNI likes a -polort 4 for this, so I oblige. I get nice activation by Event B contrasted with Event A.
However, I want to dump out trial-type-averaged signal change data elicited by certain event classes, using a mapfile applied to the time series data. The problem is, that event classes are not evenly occurring in time due to the results of pseudo-randomization of event timing, and visual inspection of the time series in most voxels indicates various drifts etc (scanner warming up, head crushing cushion-- whatever). The result is, that if I apply a mapfile to the raw dataset, with averaging of duplicate events, I get really weird results.
I am interested in suggestions as to what would be a a kosher way to pre-process the time series data to minimize the impact of low-frequency shifts or linear-ish trends in signal (i.e. that have nothing to do with the events of the task)-- BEFORE applying the map file.
One idea might be to simply use 3dFourier to apply a highpass filter at the end of preprocessing. Would this be kosher? If so what -highpass cutoff value would smooth the signal roughly analagous to accomodating it by MODELING the baseline polort 4?
