Note that removing the lowest frequencies is essentially
what the -polort regressors are for in a normal execution
of 3dDeconvolve.
For a task based connectivity analysis between conditions,
do you have long blocks for each condition, or is it a
fast, event related design (or somewhere in the middle)?
So I gather the questionable step is leaving the BOLD
response from tasks of interest in the data when computing
the correlations. Is that right?
Censoring before bandpassing breaks the time axis in a
model that is all about signal frequencies. Censoring
after bandpassing can send echoes of would-be-censored
spikes ringing through a time series.
If you want to slip bandpassing in there, the place to do
it is in the regression (via 3dDeconvolve or 3dTproject).
Censoring and bandpassing should be done in the same step.
That means no calls to 3dBandpass.
It seems to me that you could easily carry out your test
exactly like afni_proc.py would do it: add bandpassing to
the 3dDeconvolve command, and possibly then use 3dTproject
with the resulting X-matrix, since it is much faster.
Why don't you just write an afni_proc.py command, and copy
the bandpassing bits into your own script? At least look
it over to understand how it is done (and done efficiently).
It is still not clear to me what would be wrong with a
normal afni_proc.py script.
- rick