Dear Shary,
There are a number of ways to test for a change in scan data, in relation
to a change
in a psychological measure. The simplest way is to subtract the two scans
and perform
a linear regression analysis using the change in MMSE score. To avoid
masking and
proportional scaling outside SPM it may be simpler for you to use the model
below
and test for the time x MMSE interaction with a contrast
0 0 -1 1 (or 0 0 1 -1)
In your current model the main effect of time (assuming you have mean
corrected
the MMSE1 and MMSE2 scores) is
1 -1 0 0 (or -1 1 0 0)
The main effect of MMSE is
0 0 1 1 (or 0 0 -1 -1)
and the interaction is the change in SPECT that depends on MMSE score.
This is not
the same as a correlation between changes (in SPECT and MMSE). To look for
this association
you would have to model these changes explicitly as described above.
I hope this helps - Karl
____________________________________________________________________________
_______________
Dear Karl,
We have a group of patients who are treated for 12 months. Patients
received SPECT (SPECT 1) and MMSE (MMSE1) before treatment and SPECT
(SPECT2) and MMSE (MMSE2)after treatment. MMSE is a clinical measure
(minimental state examination score), which will be taken as covariate.
Will the following contrasts + interpretations be correct if I do
multisubject conditions and covariate to see the correlation between SPECT
changes with MMSE changes:
Contrasts:
SPECT1 SPECT2 MMSE1 MMSE2
-1 1 -1 1
The above shows the regional perfusion increases which are correlated with
MMSE score increases. In other words, regional perfusion increases with
increase in MMSE scores.
SPECT1 SPECT2 MMSE1 MMSE2
1 -1 1 -1
The above shows the regional perfusion decreases which are correlated with
MMSE score decreases. In other words, regional perfusion decreases with
decrease in MMSE scores.
|