Hi Emily,
Scrubbing is just censoring, I think they were simply not aware
of it when they made up the word scrubbing.
The metrics they used for censoring are similar to what afni_proc.py
uses (the enorm, though one can also use outliers). Powers used
FD and DVARS.
afni_proc.py enorm: take motion derivatives (first difference), and then
the per-TR Euclidean norm: sqrt(sum squares).
afni_proc.py outliers: fraction of brain voxels that have outliers at that
point (see 3dToutcount -help for a description).
Powers FD: take motion derivatives (similar to enorm), but instead of
sqrt(sum squares) use sum(absolute values). Also, they scale the
rotations down slightly. I prefer the enorm to this.
Powers DVARS: (from what I recall) this is the same as AP's enorm,
except that while afni_proc.py uses the motion parameters, this
comes from the actual EPI time series. I like this metric, but have
not yet bothered to implement it in afni_proc.py (partially since no
one has asked for it).
Anyway, the main point of censoring with these metrics is to exclude
high-motion TRs from the regression. There are surely many good
metrics for doing so.
- rick