Dear Gordon, Rich and SPM:

Warning- long winded answer...

The slice timing help text certainly agrees with this point of view, but I think
only discusses part of the story. I agree that slice timing should be performed
before realignment in the case where there is little subject movement within
run. However, Andersson, Henson and others have also cogently argued that it
might be better to perform slice timing after realignment when there are
"larger" movements. The argument is along the lines of: let's say you have a
voxel which is supposed to be in cortex; in scan volume x you get appropriate
MRI signal from cortex, but in scan volume x+1 the subject moves and the signal
comes from something outside the brain. In this case slice timing correction
will spread that bad signal into that voxel's time series. However, in this
scenario slice-tming correction after realignment would probably include signal
for that voxel from a more similar brain region. The issue can be complicated
considerably by motion within plane vs. motion between planes particularly in
the case of an interleaved acquisition.

I don't have a better solution other than performing slice timing before
realignment when movements are small within run, and after realignment when
movements are larger. Sometimes this may be clear, for example, when there are
separate populations- e.g., patient v. non-patient.

The decision as to what is a small vs. large movement is left as an exercise for
the reader. :) Perhaps someone has done some work on this - and searching the
spm list will return multiple comments about this issue.

Regards,
Darren



Quoting Rich Hammett <[log in to unmask]>:

> Dr. Gordon D. Waiter wrote:
>
> > Having processed a data set twice, once with slice timing before and
> > once with slice timing after realignment and noticed different results
> > at the results stage, is it best to do slice timing before or after
> > realignment in a descending acquisition. If there is a preferred order
> > could someone explain why.
>
> Slice-timing correction is ONLY valid if it is applied before the actual
> data is changed between slices in any way.  Registration, smoothing,
> realignment all move data from one slice to another.  If SPM actually
> modifies the data in these steps, there is no way to keep track of
> which data was in which slice initially.  One of the SPM gurus will
> have to verify that.
>
> I can't see any good reason to do slice timing correction AFTER your
> first step.  Now, deciding whether the slice-timing correction is
> meaningful and helpful is an entire other story...
>
> rich
> --
>
> \  Rich Hammett       http://afni.nimh.nih.gov
> /  [log in to unmask]          sun dogs fire on the horizon
> \                      meteor rain stars across the night
> /                                this moment may be brief
> \                                 but it can be so bright
>