Dear experts,
I am wondering how I might do a conjunction analysis in SPM8 from 2 separate
models.
SPM only allows one to select contrasts for the conjunction analysis from the
same SPM model. Is there any workaround - to select 2 contrasts from
separate models?
Are there any plans to allow this in the future?
Or allow one to use implicit masking across models? Which doesn't do the
same thing but might help if a conjunction across models isn't possible.
I have been reading the posts on the list and it seems a possible way to do a
conjunction across models may be to save separate t-maps from each model.
I could then use imcalc to create a new image (using the & function). This
would just give me a new image that I could superimpose on a template and
visually inspect where the areas specified in the conjunction are. Is there a
way of finding out what the MNI co-ordinates of these areas are and what the
p-values might be?
The reason that I am trying to do this is that I have 2 groups on which I am
doing a VBM analysis.
I want to compare them and covary e.g. for age.
I also have another variable of interest - score.
I want to see what areas are atrophied in one group, what areas deteriorate
when low scores are produced, and lastly what areas deteriorate that are
related to lower scores (the conjunction).
I can see that I could set everything up as a single model. E.g. If I do a two
sample t-test I could enter covariates (putting both age and score here). If
the variables were inputted in this order and I put 1 -1 0 0 as a contrast, this
would show me the areas that differ between groups (in the specified order)
and would covary for age and score. I could then make a contrast for score
e.g. 0 0 0 1. If I tried the conjunction between grp1 vs grp 2, and the
positive correlation of score (by selecting both contrasts using the ctrl
button) - surely this would not be informative as initially the group comparison
contrast would be calculated covarying for score. Therefore the conjunction
between the score contrast and the group difference contrast should not be
significant in any brain region. The reason being that score was covaried for.
Is that correct?
Should I not have 2 separate models in which I have a two sample t-test with
both groups entered and only age as a covariate (so as not to covary for
what I am interested in - score), and another regression model where I would
have age and score. With this method I would need a way to select
conjunction across models.
Hopefully someone can help.
Thanks in advance.
Will
|