Dear Rachael,
I may have misunderstood what you are trying to do, but since noone
else has offered you an answer yet I thought I give it a shot.
>I'm still struggling a little with understanding how contrasts need
>to be specified in spm99... The timecourse does not always look like
>I expect it to!
>I have the following blocked fmri design:
>control task control task control task
>If a set it up as one condition and use the contrast [1] then I see
>areas that are activated with respect to the control. The contrast
>[-1] reveals areas that are deactivated with respect to control.
So far so good.
>However, I'm interested in the task vs. control (activation) x time
>interaction. So I specified the design as 3 conditions. I thought
>that I could mask [1 1 1] (task activated wrt control) with [1 2 3]
>(increasing trend over time) or [3 2 1] (decreasing trend over
>time). However, the [1 1 1] contrast appears to show regions that
>are both activated and deactivated for task wrt control. Can anyone
>explain this?
I don't quite understand what your three conditions were. Perhaps
you tried covariate 1 for the task condition, covariate 2 for the
control condition, and covariate 3 for the condition by time
interaction term. In this case your model is overspecified. Just as
in the first case above, you only need to model the task vs control
comparison once.
I would have expected that you have a model in which covariate 1 is
the task vs control box-car (probably convolved with the hrf; that's
what most people would do), and covariate 2 is the condition by time
interaction. You would do this by specifying that you have 1
condition, but then asking for 'time' when SPM asks if you want any
parametric modulation. This now adds a second covariate to the
design matrix in which, at the start, task conditions appear in white
(i.e. are modelled by +1s), but by the end of the time series they
are appear in black (i.e. modelled by -1s). All of the control
conditions are modelled as mid-grey (i.e. zeros). As always, SPM
would add a third covariate which is just a column of +1s.
Now if you are interested in voxels which show a significant main
effect of task (vs control) then the contrast is 1 0 0 as before.
However, if you want voxels which show a significant
condition-by-time interaction, then you use contrast 0 1 0,
Best wishes,
Richard.
--
from: Dr Richard Perry,
Clinical Lecturer, Wellcome Department of Cognitive Neurology, Darwin
Building, University College London, Gower Street, London WC1E 6BT.
Tel: 0207 679 2187; e mail: [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|