In a recent study, we experimented with using:
a) the 6 motion estimates
b) the 6 motion estimate derivatives
c) the RMS value of the 6 motion estimates
When evaluated in terms of change to the final t-statistics, all methods
proved pretty much equivalent. This was for both a block design and a
rapid event-related design, evaluated for a group of 33 subjects. In the
end, we just reported one of the above measures. The paper is in press
in Human Brain Mapping:
Johnstone, T., Ores Walsh, K. S., Greischar, L. L., Alexander, A. L.,
Fox, A. S., Davidson, R. J., & Oakes, T. R. Motion correction and the
use of Motion Covariates in Multiple-Subject fMRI analysis. Human Brain
Mapping, in press.
More recently, we have found that the motion can produce effects that
are quite delayed with respect to the motion itself, so we now include a
1-TR lagged motion estimate as a regressor in addition to the in-synch
estimate.
With respect to displacement not being uniform across the brain: The
BAMM software sounds interesting, and it would be preferable to
effectively correct the problem of motion rather than simply try to
statistically account for it. That said, when using motion covariates
the absolute magnitude of estimated motion at each voxel is not really
relevant - for each voxel a different parametric fit will be calculated.
So it's the relative displacement for one time point compared to a
reference for a given voxel that is important - if this scales
relatively linearly with the overall volume motion estimates, then the
covariates will do a reasonably job. Of course, this might not be the case.
Tom
----- Original Message -----
From: Alle Meije Wink <[log in to unmask]>
Date: Wednesday, October 5, 2005 7:02 am
Subject: Re: [SPM] motion parameters as covariates
> Torben Ellegaard Lund wrote:
>
> > In Karl's MRM paper on residual movement artefacts: Magn Reson Med.
> > 1996 Mar;35(3):346-55. Movement-related effects in fMRI time-series.
> >
>
http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=pubmed&dopt=Abstract&list_uids=8699946&query_hl=1>
> > they use an expansion of the movement parameters. The expansion
> > includes in addition to the 6 rigid body movement parameters for a
> > volume acquired at time t: R(t) it also the movement parameters from
> > the previous volume R(t-1) and these terms squared: R(t).^2 and
> > R(t-1).^2, this is 24 extra parameters in your design matrix. In
> this> way you can model both linear and quadratic term of both the
> movement,> but also of the velocity of the movement as well as
> history effects.
>
> A combination of these methods would probably be optimal: compute the
> displacements D for each voxel, and use D(t), D(t-1), D(t).^2, and
> D(t-1).^2. That would model linear and quadratic displacement
> terms, as
> well as velocity and history effects. Computing the displacements as
> scalars instead of using the 6 parameters, does not introduce 5
> `virtual' degrees of freedom for each of those terms, which makes the
> analysis more powerful.
>
> Best wishes,
> Alle Meije
>
>
> ___________________________________________________________
> To help you stay safe and secure online, we've developed the all
> new Yahoo! Security Centre. http://uk.security.yahoo.com
>
|