Hi,
I understand the difficulty here as we also look at patients where
it is very difficult to eliminate movement. I have personally been
looking at a way of making regressors to remove the most
substantial motion effects and have had some success. It is a
little different from what you describe and I would be happy for
you to try it out and see how it works for you. If you are interested
in this just let me know by email off the list.
With respect to your more specific question: you certainly do not
want to be changing your stimulation EVs like this, due to the
HRF convolution as you say. You also do not want to convolve
your motion EV, but you do want to have temporal filtering act
on it (since this is also happening to the data you collect). As
for orthogonalisation - you do not need to worry about this as
long as your contrasts of interest do not include your motion
EVs. If that is the case then orthogonalising your stimulus
wrt the motion will give you exactly the same answers as
not orthogonalising, for those contrasts only using the
stimulation EVs. This is generally true for any confounds,
including the simple motion parameters. So the bottom line
is that you don't have to worry about this as you are already
getting what you want (anything looking like either motion
or stimulus gets interpreted as motion and not stimulus
which is the conservative approach).
All the best,
Mark
On 31 Mar 2008, at 23:47, Jennifer Bramen wrote:
> Dear All,
>
> We work with children, who tend to move a lot, Becasue we also work
> in a
> difficult torecruoit clinical population, we keep data which may
> otherwise
> have too much motion for other research projects to consider
> including. To
> deal with this, the lab has tried to find ways of completely
> regressing out
> time points with excessive motion (movement of more than 2 mm).
> First, a
> motion EV is created, where any TR with excessive motion is assigned
> a 1 and
> all others a 0. This EV is of course not convolved with anything.
> However,
> this alone could potentially leave some effect of motion in the
> data. To
> deal with that, the lab has been editing the stimulus EVs by placing
> 0s in
> TRs where there is a corresponding 1 in the motion regressor. This
> we now
> believe is problematic as these 0s are convolved with an HRF, having
> a long
> lasting impact on the model, where motion should only have a
> discrete effect
> on the data. Instead, would orthogonalizing the stimuli wrt EVmotion
> do what
> we want?
>
> Thanks
>
|