You may also want to reduce the 6 components to a fewer number of
orthogonal components with SVD/PCA. Typically, the 6 waveforms are
highly correlated, which can make your design matrix ill-conditioned. In
theory, this is not a problem because you only want to remove the
subspace spanned by the 6, but in practice it might cause some problems.
doug
Vikas Rao wrote:
>
> Has anyone successfully used motion in a model as a covariate of
> non-interest? I've managed this with other software packages (VoxBo)
> but I am getting strange results with FSL. Adding motion as a
> covariate of non-interest didn't change the results at all, while there
> is a huge motion-related spike in the timeseries. Basically what I did
> was load the prefiltered_func_data_mcf_abs file as a 1-column vector,
> not convolved, and orthogonalized w.r.t. all covariates of interest.
>
> If you've got this to work, did you :
>
> A) Use the absolute motion, or put in a covariate for each of the
> directions of motion?
> B) Did you renormalize the motion param eters at all before feeding
> them into FSL (for example demeaning them)?
>
> Thanks,
>
> Vikas
>
> ****
> Vikas Rao
> Lab Assistant
> [log in to unmask]
>
> Clinical Neuroscience Lab, UCLA
> www.cannonlab.net
> 310-794-9673
--
Douglas N. Greve, Ph.D.
MGH-NMR Center
[log in to unmask]
Phone Number: 617-724-2358
Fax: 617-726-7422
|