Hi Ajay,
The the 'N' stands for noise, so it seemed appropriate
to compute it from the noise time series, when all
known signals have been regressed out. I believe it
is often computed from non-brain, non-ghost areas for
the same reason, but that is much harder to automate.
In particular, it did not seem reasonable to include the
correctable motion effects, or even BOLD signals of
interest.
There is an -volreg_compute_tsnr option, in case one
wants to compute it from the volreg output, too.
- rick