Dear Chung-Haow, Andrew,
Just to put in a belated reply here:
> At 19:50 27/05/99 +0800, Chung-Haow Tu wrote:
> | According to the Dr. Brett's abstract,
> | the scan order effect looks like a movement artifact. So SPM
> | will include this in motion correction in feature?
>
> I doubt it.
> (We're talking PET/SPECT here, not fMRI, which has motion correction.)
>
> PET/SPECT analyses can have very low degrees of freedom. Assuming motion is
> simply a nuisance effect (not confounded with the effect of interest),
> inclusion of motion parameters into the statistical model could potentially
> have a detrimental effect, by using up more in degrees of freedom than is
> gained through a better fitting model. This is therefore a model selection
> issue. Thus, inclusion of such covariates needs to be left as a user
> choice.
I do agree that it would not be wise to include the motion parameters in a
PET design by default, if there are low degrees of freedom. However there
does seem to be a strong relation of PET signal in some areas to motion
parameters in (all) the activation data that I have looked at.
Thus, if there are reasonable degrees of freedom, and/or if there is apparent
signal in the inferior frontal areas (see the pictures in my abstract) then
I would suggest that it would be wise to assess the effect of adding both scan
order and motion parameters to the PET model.
> Additionally, there may be alternative pre-processing steps that can remove
> some of the artefact referred to by Matthew...
I have tried and failed to do this, and it is not straightforward. So,
until there is another accepted method, I think we have to use the statistics,
as for fMRI.
Cheers,
Matthew
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|