I have a question about the use of ANCOVA in the design matrix for a PET study.  I have read several emails regarding this, but I am still unsure about it.

I have an FDG-PET study with 30 normals and 26 experimental subjects.  Each subject has a single scan.  The MRI and PET are coregistered.  The normalization parameter file from the MRI segmentation is used to warp the PET data.  The global mean of the experimental subjects is approximately 6% lower than the controls (as expected), however, the subject population has a good degree of variance.  I wanted to use global normalization to remove the variance in each group due to variable injection dosage.  Since the means are not the same, I need to normalize the globals from each group independently.

It appears that the ANCOVA option in the design is appropriate.  In other emails it seems like this is used in activation studies rather than single scan FDG-PET. Can someone provide insight into whether this is the correct usage of the ANCOVA option, as it has a dramatic influence on the statistics.

Also, there are two other options that I need clarified: Overall grand mean scaling (I assume this is supposed to be no); and Normalization - I assumed that the ANCOVA in the design matrix will remove the effects of the globals, so do I still need to normalize the scans - and if so, should I use ANCOVA or proportional scaling?  When I perform the statistics, I am not selecting the ANCOVA global regressors in my contrast because I want to remove these effects - is this correct?

Thank you in advance for any help that can be provided.

-Jeff Stout