Hi Michael,
3dSeg seems to be complaining about NaN values in
the dataset. Most AFNI programs were not written
to process that in some special way, so I suggest
that you zero out those values ahead of time.
Consider using @float_fix for this.
I am not sure whether 3dSeg has an option to skip
its internal bias field correction, but even if so,
that would probably just postpone NaN issues until
later.
- rick