Dear SPM list
I'd be grateful if someone could check my understanding and
implementation of the following PET/SPECT model and contrasts in SPM99.
If I've gone wrong, or if there is a better way to do it, please let me
know. Thanks in advance.
I have 16 subjects, 2 conditions (1 pharmacological intervention and 1
control) and 1 SPECT scan per condition measuring rCBF. In addition I
have a behavioural rating (in this case subjective mood) for all
subjects at the time of each scan. My intention was to look at group
changes in rCBF between the 2 conditions, and when I selected design
type 'Population main effect: 2 cond's, 1 scan/cond (paired t-test)' I
found changes in anatomically sensible regions. However, I have also
(unexpectedly) found a significant change in mood between the two
conditions.
I therefore wish to find regions in which changes in rCBF are associated
with the conditions when any association between mood and rCBF has been
removed. And (since I'm doing that) to have a look at the regions in
which mood and rCBF covary.
So, I have selected:
Model: Multi-subj: conditions & covariates
#subjects: 16
Entered all the scans in the order 1. Pharmacological intervention 2.
control
# covariates: 1
[32] covariate: I have entered the 32 raw (not mean corrected) mood
scores in the order - Subj1: Pharmacological intervention,
Subj1:Control, Subj2:Pharmacological intervention,
Subj2:Control,........Subj16:Pharmacological intervention,
Subj16:Control
Covariate interaction: none
Covariate centre: around overall mean
# nuisance variables: 0
proportional scaling
PropSca global mean to: 50
analysis thresh (prop'n of global): 0.8
mean voxel value (within per image fullmean/8 mask)
I end up with a design matrix with 2 columns for the conditions, 1
column for the covariate and 16 columns for the subjects.
So, first of all, are all the above the most appropriate choices? Am I
right to covariate-centre around the overall mean or should I choose
'none'? And coincidentally, what is the difference between what I am
measuring above and what I would be measuring if I chose covariate
interaction by subject (provided I had more scans per subject) or
covariate interaction by condition?
Moving on to the contrasts - am I right to think that the following are
correct?:
(1 -1 0) will show regions for which rCBF is greater in the
Pharmacological intervention after the effect of mood on rCBF has been
removed
(-1 1 0) will show regions for which rCBF is greater in the Control
condition after the effect of mood on rCBF has been removed
(0 0 1) will show regions in which there is a positive correlation
between mood score and rCBF. This will include regions in which
increasing mood score is associated with increasing rCBF and in which
decreasing mood score is associated with decreasing rCBF.
(0 0 -1) will show regions in which there is a negative correlation
between mood score and rCBF. This will include regions in which
increasing mood score is associated with decreasing rCBF and in which
decreasing mood score is associated with increasing rCBF.
(Assuming the above are correct), in relation to (0 0 1) and (0 0 -1)
above, what further contrasts or maskings would enable me to separate
out the areas of increased from the areas of decreased rCBF in these
contrasts?
Thanks so much for your time and advice.
Peter
--
Dr Peter Talbot MB ChB MRCPsych
Postdoctoral Research Fellow
Division of Functional Brain Mapping
New York State Psychiatric Institute
Unit 31, 1051 Riverside Drive
NY 10032 USA
Tel 212-543-5975
Tel 212-543-5040
Fax 212-568-6171
Email [log in to unmask]
_____________________________
|