Hi Mehul,
just my 2 cents:
> However I am a little apprehensive about that finding given our
> failure to demonstrate MTL activation when motion parameters were
> included. Obviously, the results seem to indicate that MTL activation
> correlated with motion.
This is not necessarily true: when including the motion parameters, you
do not only explain variance induced by motion but you also loose 6
degrees of freedom. Even if you include 6 nonsense parameters (I tried
random numbers for fun a while ago) you may weaken your main effect
substantially, depending on how robust it is and how many observations
you have in the first place. When converting into a t-value (and
deriving a p), your fewer degrees of freedom may hurt you even in the
absence of variance meaningfully explained by the regressor. You can
actually see where and how much variance is explained by the motion
parameters by doing an f-test on them.
> I've read several papers and there appears to be no consensus about
> whether or not to include motion parameters in the 1st level design
> matrix. Even within the same lab, I've seen papers published both
> with and without movement parameters included in the design matrix.
... for which there may be a good reason: for motor paradigms (or in
patients), typically involving more task-related motion, it may be
meaningful, while for a silent cognitive design (or in a group of
healthy subjects), it may not be. In the absence of relevant motion (see
also below), it may not be either.
> I've searched through the listserv with no luck, it seems like most
> of the relevant posts end with some version of "you should include
> the movement parameters, except when your activation in key brain
> regions is correlated with motion." Unfortunately, that doesn't
> really help me.
One way to get a feeling for this is to calculate the correlation
coefficient between your motion regressors and your task offline. This
is very rough but it does give you an idea on how big the problem is.
> I guess what I'm looking for is some guidance about how to proceed,
> how would I justify not including the movement parameters in the
> design? are there any alternatives to including the movement
> parameters in the design matrix? because I have not come across any
> on the listserv
Including them is not always helpful, so, as with every other step in
data processing, you need to have a good reason for it. I believe in
throwing out bad data, so setting an absolute threshold of tolerable
motion (typically, 1 voxel size is used, although I have yet to find the
original source for this rule-of-thumb) is always a good idea. If you
have a clean dataset (say, maximum motion is always below 1mm), I would
see no need to include the parameters. An alternative is to use realign
and unwarp, which tries to model some of the motion effects (motion*B0
interactions) and may be an option for you. As per Jesper, using both
realign & unwarp and including the motion parameters is overkill, so
using the one or the other in the presence of motion may be worth a try.
Hope this helps,
Marko
--
=====================================================================
Marko Wilke (Dr.med./M.D.)
[log in to unmask]
Universitäts-Kinderklinik University Children's Hospital
Abt. III (Neuropädiatrie) Dept. III (Pediatric neurology)
Hoppe-Seyler-Str. 1, D - 72076 Tübingen
Tel.: (+49) 07071 29-83416 Fax: (+49) 07071 29-5473
=====================================================================
|