Dear Meredith,
I don't have any definitive answers but I'll give these a shot...
I'm pretty skeptical about removing volumes and replacing them with a mean
of the temporally adjacent scans -- particularly in any event-related
design. I suppose in an extreme case it might be ok to do it but if I had
to replace more than about 1 volume in 100 then I almost certainly wouldn't
use this procedure. If you're replacing that many volumes my guess would
be that there is some serious acquisition problem (drop-out, scanner
instability, bad motion, etc) and personally I'd throw the data set out and
try to fix the basic problem in acquisition rather than at the
post-processing stage. This is a fairly conservative approach but probably
a safe one.
I use the same approach for correlated head motion. If the head motion is
significantly correlated with the stimulus paradigm, I throw the data set
out as there is no way to unconfound these factors in the GLM. If the
motion is really small (e.g. much less than the minimum voxel dimension)
then you probably don't need to worry about it anyway, and in that case I'd
just do realignment ("motion correction") and proceed normally. Jesper
Andersen has a really nice tool for correcting motion x susceptibility
interactions which can be very useful for stimulus correlated motion. For
the moment it is implemented as a toolbox in SPM2 but you can use it for
realignment and take the results back to FSL, if you want to use it.
Hope this is some help.
Joe
--------------------
Joseph T. Devlin, Ph. D.
FMRIB Centre, Dept. of Clinical Neurology
University of Oxford
John Radcliffe Hospital
Headley Way, Headington
Oxford OX3 9DU
Phone: 01865 222 738
Email: [log in to unmask]
|