Tobias -
> I have a question concerning the correction for movement artifacts using the
> realignment parameters.
>
> We have several fMRI-datasets which show deviations from the reference image over
> the time course of the experiment, as can be seen in the "ps"-file graphs after
> the realignment procedure.
>
> However, those deviations are considerably different from subject to subject.
> They range from a maximum of about 1.5 mm in one subject to about 5 mm in another
> subject. Unfortunately, we are not able to completely eliminate head movements by
> better fixation of the head so that we just have to deal with the movements.
>
> Now my questions are:
>
> Is there any rule of thumb as to when to include the realignment parameters in
> the design matrix ?
If the movement parameters were uncorrelated with your effects of
interest,
their inclusion in your model can produce a significant improvement in
your
model fit and results (by accounting for unwanted additional variance).
The danger comes when movement is correlated with your effects of
interest
(eg head movement each time a stimulus is presented). Including the
movement
parameters in this case will reduce your power in detecting the effects
of
interest. While you could orthogonalise the movement parameters with
respect
to your effects of interest, still potentially improving overall model
fit,
there is no way of knowing whether the shared variance (that is
effectively
attributed to your effects of interest in this case) really reflects
signal
changes owing to your effects of interest, or simply movement artifact.
I prefer to err on the side of caution, and include movement parameters
(without orthogonalisation). While I might miss real activations
that happen to be correlated with movement (ie type II errors), I view
this as less serious than making false positives owing to movement
artifact (ie type I errors). In my experience, their inclusion often
actually improves model fits and my results (by soaking up residual
error).
> And if I decide to include them, do I always have to use all six of them or just
> the ones exceeding some criteria, in other words showing the greatest
> translations ?
I use all 6, because overall df's are normally high in a Fixed Effects
model
with multiple scans (and these df's are irrelevant to subsequent Random
Effect analyses). If you wanted a more parsimonious model, you could
include just the first few principal components of the movement
parameters.
> Moreover, if I include them for one subject, don't I have to include them
> automatically for all others, since, as far as my understanding goes, those
> additional covariates change the overall model ?
I would include them for all subjects, for consistency (and so that the
subject models are more "balanced" for subsequent Random Effects
analysis).
As for affecting the model, it depends what you mean by "overall model".
If you are referring to a group, Fixed Effects model, I would include
movement parameters as separate columns for each subject (the default
case
if entered as user-specified regressors in SPM). In this case, while
they do
affect "overall model" fit in terms of residual error, they do not
affect
the parameter estimates for other subjects.
> I analyzed several datasets both with and without inclusion of the realignment
> parameters and got considerably different results for some of them.
> So, I am not sure about how to proceed because of that.
To be expected, since they change the residual error and might be
correlated
with your effects of interest (see above). You could test the
correlation
between your covariates of interest and the movement covariates (check
the
orthogonality matrix for example).
Rik
--
---------------------------8-{)}-------------------------
DR R HENSON
Institute of Cognitive Neuroscience &
Wellcome Department of Cognitive Neurology
17 Queen Square
London, WC1N 3AR
England
EMAIL: [log in to unmask]
URL: http://www.psychol.ucl.ac.uk/rik.henson/index.html
TEL1 +44 (0)20 7679 1131
TEL2 +44 (0)20 7833 7472
FAX +44 (0)20 7813 1420
---------------------------------------------------------
--
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|