Hi charles,
> I need some help about how to specify a VBM design matrix.
>
> We have got two groups, patients and control subjects, aged from 10 to
> 15 years, paired in age and sex, two scans per subject. In the group
> of patients the first scan was pre-treatment and the second one once
> it was finished. The delay between scans was from 6 to 8 months.
So, if I understand correctly, you have 4 conditions Controls T1,
Controls T2, Patient T1 and Patient T2 with a drug effect only in patients.
> We are looking for those differences in patient group that are just
> due to treatment, that is, substracting the possible effect of
> growing, using the GS volume, the age and the number of days between
> scans as a confounds and taking into account that patients and control
> subjects were paired in age and sex. We also would like to know if
> those differences are correlated or not with the result of a
> neuropsicological test.
>
A multiple regression analysis should be able to do the job: 4 groups
and 1 column per variables (age, volumes, ...) and next contrast groups
like C1C2 vs P1P2, C1P1 vs C2P2 or C1P2 vs C2P1. To investigate the
correlation between GM and neuropsy scores, you have two different ways
to specify the design matrix. With the design mentioned here above, you
do not distinguish groups along the variable of interest, and then you
can demonstrate a linear correlation along this dimension in both
controls and patients with, for example, high scores in controls for
high volumes and conversely low scores for small volumes (simple
contrasts +1 or -1). The other way is to distinguish groups along such
variables with several columns witch allows to compare regression
coefficients, for example in this part of the brain the volume is
positively correlated the scores in controls whereas no correlations
were found in patients (contrast 1 -1 or more complex). For this last
possibility, you need of course a 'large' sample, as you will then add
columns, i.e. you remove df.
Hope it helps,
Best
- cyril
|