| With the aim to differences compare age-related changes between
| functional (perfusion) and structural (grey matter) data (cfr.
| Richardson et al., Brain 1997, 1961-73), smoothed to same resolution,
| I have used the following design :
|
| Single subject:conditions and covariates with 2 conditions (1
| functional, 1 structural), 1 covariate(age), ANCOVA with global
| scaling to overall grand mean (= 1 nuisance variable).
I think proportional scaling should be a better model. What it
does is approximate perfusion/grey_matter changes by
perfusion-grey_matter changes. For this to be a reasonable
approximation, it is necessary for all the images to be scaled
the same.
As a co-author, I do have some reservations about the Richardson
et al. approach, but I think it can be a reasonable first pass analysis.
See also:
http://www.mailbase.ac.uk/lists/spm/2000-07/0185.html
http://www.mailbase.ac.uk/lists/spm/1999-10/0190.html
|
| The design matrix shows 'cond0', 'cond1', 'age@condition0',
| 'age@condition1', 'mu' and 'global parameters'.
|
| Is such design appropriate and is it correct to interpret the
| contrast 0 0 -1 1 as determining regions where age-related decreases
| in the functional data are larger than age-related decreases in grey
| matter concentration ?
This is how the results would be interpreted, although they are only
really meaningful if the two sets of data are scaled the same. E.g.,
Consider a voxel where perfusion falls from 0.05 to 0.0, and grey matter
concentration falls from 0.5 to 0.45. This would not show any
interraction in the SPM analysis, but if you consider what happens in
terms of perfusion per unit of grey matter, then there is something
significant going on.
Best regards,
-John
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|