Dear list, sorry for the repost, but I would really like to have some tips on this one.
Dear SPM list,
We're working on PET-AV45 data right now. We are trying to create the so-called W map. A W map is a map that show the effect of a variable (say, pathology) on a certain imaging modality (say, PET-AV45) cleaning out the effect of another variable (say, the age).
To do so, theoretically, once should calculate the correlation coefficient between age and AV45 uptake in controls, then calculate the SD of the residues in the controls. Finally, the W map is given by the AV45 map of the patient minus the predicted AV45 map of that patients on the basis of the patient's age, divided for the SD of the residues in the controls.
From the theoretic point of view, I got no problem.
When I try to step in to the practical part, the problems arise. In particular, in order to obtain the SD of the residues in the controls, I performed a regression in SPM using age as covariate. Having just one variable in the design matrix, I should have beta_001 = correlation coefficient and beta_002 = intercept, am I right till here ?
Then, to calculate the residues of each subject, I simply implement, , the following formula: AV45 image -(beta_001*age (demeaned)+beta_002). Where beta_001*age+beta_002 it's simply the image predicted on the basis of the age of the subject.
The problem here is that while the values on the AV45 images are in a range of 0-4 (is a ratio calculated using the cerebellum as reference), the values in the beta_002 image are in a range of 10-50. This means that my residues are *always* completely out of range.
Another problem is the demeaning: while I understand *why* I have to demean, I don't know now how to deal with negative value, because, conceptually I can't have a negative uptake value. Should I just use the module of the demeaned age ?
I hope that you get to the end of this wall of text and that you can give me some tips.
Thank in advance
F.
|