Tetsuya,
> In my experience as a beginner of fMRI study, some deactivation
> ( i.e. activation in rest minus task contrast) may dissapear after
> reanalyzing the same dataset with "no global normalisation".
> [snip]
> Which of ANCOVA or scaling or no gobal normalisation is appropriate
> for fMRI data?
This is a somewhat tricky question. We recently published a paper
(Aguirre GK et al. 1998, NeuroImage 8, 302-306.) arguing that correcting for
the global signal (either by ANCOVA or scaling) is undesirable when the
global signal itself is correlated with the task covariates of interest.
Because of the alterations you see in your statistical maps dependent upon
the inclusion/exclusion of the global signal, I imagine that such a
correlation exists in your data set.
So what does it mean if there is a correlation between the global signal
and the covariates of interest? The following is from the paper I mentioned
before:
It seems that when global signal covariates are included within an analysis,
supra-threshold voxels should be regarded as those with a relationship that
is significantly greater than that which the global signal enjoys.
Furthermore, when a global signal covariate is present, it is inappropriate
to automatically infer that regions with a negative relationship with the
task have decreased signal values during control compared to experimental
periods (for example), as this statistical circumstance may arise if the
signal in a voxel is simply less correlated with the task than is the global
signal itself.
Hope this helps,
Geoff
--
Geoffrey Karl Aguirre, Ph.D.
Univ. of Pennsylvania
(215) 614-1976
mailto:[log in to unmask]
http://cortex.med.upenn.edu/~aguirre
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|