I'll have a stab at this.
Neuroimaging involves digital sampling and fMRI images have rather course spatial sampling resolution.
Resampling during realignment involves some form of interpolation and that will introduce errors - think about partial volume effects.
The realignment procedure uses an iterative procedure to minimize the error between a template image and the target image.
By definition, the minimizing of the error means that some residual misalignment error may exist. Modeling the motion parameters as regressors, while reducing a dF for each, *should* help in reducing the error term in the linear regression model. You could try with and without to see if there's any noticeable improvement.
Cheers - Mike
>-----Original Message-----
>From: SPM (Statistical Parametric Mapping)
>[mailto:[log in to unmask]] On Behalf Of Markus Burgmer
>Sent: Tuesday, January 30, 2007 6:41 AM
>To: [log in to unmask]
>Subject: [SPM] motion correction
>
>dear spmers.
>
>i searched the archives about a question about motion correction and
>covariate for the model, but did not find a sufficient answer
>(some were to
>technical for me to understand, sorry about that).
>i always understood realignement in spm in the way, that
>afterwards (after
>reslicing the images) spm corrected for the motion between the
>images. for
>me it sounds like after that, almost no differences in motion
>between the
>images exists, the images almost "look" like one static image over the
>whole time series. (maybe here i am wrong, please correct me).
>now i read in many mails, that it is recommended to integrate the
>translation and rotation data of the rp-file as a covariate
>into the model
>to control for movement effects.
>
>now my question:
>if both of my above assumptions are correct, why should i
>integrate the
>motion covariate? didn't i control for movement artefacts by the
>realignement procedure itself? i thought that the rp-file gives me
>information about the rotation and translation which results
>in the new
>images after realignement. integrating those also in the model
>would be one
>step to much?
>
>if i still have to integrate the regressor, did i understand
>it right to do
>it in the first-level procedure?
>
>thanks in advance
>
>markus burgmer
>university münster, germany
>
|