Hi Paul,
Yes, I personally prefer using ricor and principle
component tissue regression from eroded ventricles
and white matter to bandpassing. But bandpassing
is still the norm.
There is nothing that guarantees removal of all
high-frequency noise (except zeroing all the data),
including bandpassing. And the power spectrum of a
BOLD-only time series is not all zeros above the
threshold.
The point of inefficiency is that it is does cost
so many degrees of freedom to do bandpassing.
Compare 13 ricor regressors and maybe 4 or 5 for
tissue regressors and ANATICOR with the 46 used
for bandpassing. And 46 is low.
For motion, the empirical means I tend to suggest
is based on the types of subjects one is using.
Basically, use a low threshold that still leaves
sufficient data to process. And then it is easy
to be consistent across studies once it is clear
what seems reasonable.
For "normal" adults (in resting state FMRI), one
might be able to get away with using 0.2. But
particularly since there is usually at least one
group that is more prone to motion, the number
tends to be higher.
- rick