Dear all,
I am doing a spatial regression of gray matter, white matter, and CSF masks onto an EPI 4D dataset, using the command:
fsl_glm -i <input epi data> -d <concatenated tissue probability maps> -o <output timecourse files> --demean
from which I get 3 1-dim time series largely corresponding to averaging the EPI 4D dataset within each of the tissue masks.
I noticed, to my surprise, that the output time series I get with and without the '-demean' option are *not* perfectly (to a reasonable rounding error) correlated, as can be seen from the diagonal of the following correlation matrix:
GM.demean WM.demean CSF.demean
GM 0.9788892 0.26944782 0.9140212
WM 0.7004564 0.86389531 0.4913451
CSF 0.8290315 0.01889599 0.9931620
Is this an effect of the specific numerical algorithm employed by fsl_glm? I was under the impression that centering was helpful in reducing collinearity only in the case of an interaction...
thanks in advance for any comments
very best
giuseppe
|