Hi Daniel,

|cause artifacts in the data. For example, there is a "ring" of
|deactivation in the white matter when I compare one of my event-related
|trial types to baseline. Could that be caused by motion? If so, then
|why wouldn't there be a ring around the edge of the functional image,
|too? (I'm not seeing a ring around the edge of the brain, either for
|activations or deactivations).

Yes, this might (or might not) be caused by motion. As Jesper said, the
errors introduced by motion x susceptibility interactions are likely to be
highly variable across the brain. So some areas may show artefactual
changes in signal, others may not. You can't use the presence (or
absence) of artefactual changes in one area to infer whether changes are
present in another area.

| I'm also wondering how much of this motion (i.e., 2-4 mm) is corrected
|by the realignment routine and how much "extra" motion would be
|correected by including the realignment parameters as a covariate during
|model estimation.

Unless you have some independent source of motion estimation, what you see
in SPM is what you get (corrected). The realignment routine is telling
what actually got corrected. Jesper is making the point that even if you
correctly estimate and resample ALL the motion, there may still be some
changes remaining in the resampled signal. So, for example, if
the subject shook their head violently WITHIN a scan rather than BETWEEN
scans then no amount of between-scan motion correction will solve this
problem as the motion is not apparent on comparing successive scans,
either with SPM or anything else! This is also true of the other sources
of artefact that Jesper pointed out; interpolation errors and motion x
susceptibility interactions.

| I've also heard that some people will throw out subjects whose motion
|is greater than 2-3 mm. Is that a general practice of most people who
|use SPM?

It all depends, both on the type of study (e.g. patients vs normals) and
the degree to which your particular scanner and sequence is susceptible to
motion artefacts (e.g. field inhomogeneities increase at higher field
strengths, so motion artefacts may become more prominent). There's no
single value to give out. I would have thought that if you are going to
use excessive motion as an exclusion criterion, then it should be
(a) prespecified before the experiment starts, not a 'post-hoc' throwing
out of funny looking data (b) clearly stated in the manuscript.

|problemmatic because it introduces false activations. So, for example,
|if I smooth my data to 10 mm, then motion greater than 1 mm would start
|to produce false activations and false deactivations. I'm wondering how
|people generally handle this. Do people throw out subjects whose motion
|is greater than 10% of FWHM?

I've not heard of this particular stipulation. It is clearly true that the
greater the motion, the greater the likelihood of artefactual changes in
signal. But the same caveats as above apply - no exact number is likely to
apply across all subjects, experiments, sequences and scanners.

| Also, do most SPMers use a bite bar? In my experience, motion is
|much reduced by using a bite bar. Do others find that too? How do bite
|bars and vacuum packs compare?

Bite bars are not always helpful - although they reliably return the
subject to the same position, I find them unpleasant and interfere with
subject comfort. Soft pads, or any other restraint device that tends to
return the subject to the same position if they move, are very popular and
with correct subject training and encouragement can reliably reduce motion
to 1-2mm.
 
| Also, I forgot to say earlier that I'm using an
|event-related design (1 stimulus every 16 seconds). Are event-related
|designs less susceptible to motion? I've heard that they can be, but I'm
|not sure why this should be the case.

I don't see any reason why this should be so, but maybe I'm missing
something and other list members will correct me!

| Finally, I also forgot to say that the motion in my experiment is the
|total motion across eight 6-minute runs. The data from all 8 runs is
|being analyzed together during model estimation. In any case, if a subject
|moves 4 mm, then it's usually the case that they moved 0.5 mm per 6 minute
|run rather than 4 mm in a single run. Does that make any difference? Is
|it better to have motion spread out across multiple runs than
|concentrated within a single run? I would think not, in this case, since
|I'm averaging the functional data from all 8 runs together, but I figured
|I'd ask.

I guess this is like saying 'is it better to have a small
(artefactual) activation in all four runs or a big
(artefactual) activation in just one run'. As you are averaging across
runs, the mean (artefactual) activation will be identical in each
case. But this simple-minded explanation assumes that artefactual
activation is proportional to the size of the movement, which seems
unlikely.

Best wishes,

Geraint
-----------------------------------------------------------------
Dr. Geraint Rees
Wellcome Advanced Fellow Lecturer
California Institute of Technology Institute of Neurology
Division of Biology 139-74 University College London
Pasadena 12 Queen Square
California 91125 London WC1N 3BG

voice (626) 395-2880 voice (171) 833-7472
fax (626) 796-8876 fax (171) 813-1420

http://www.klab.caltech.edu/~geraint [log in to unmask]
-----------------------------------------------------------------




%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%