> Once you define a contrast like [1 0 0 ... 0 -1], there is always a danger
> to find some brain regions which are "deactivated" by null event (or any
> other task) but actually not activated at all by the task under test. I
> have't seen any article that discusses "deactivations" by any task, but I
> expect some blood flow decrease or an increase in deoxyhemoglobin in some
> regions of the brain correlated to the null event.
This is a very complicated question!
First, I wouldn't agree that it is a 'danger' to find some brain regions
that are deactivated by null events. I think such activations might be
better characterised as data that deserve some sort of explanation! And note
that your concern is not restricted to contrasts like [1 -1] but is equally
the case with [1 0] contrasts. Almost all functional imaging involves
measuring local changes in signal from some global mean. The important thing
is I guess being very clear about *what* is being compared with *what*, in
psychological or physiological terms. I agree that 'deactivations' are
physiologically contentious, but I don't think that anyone would suggest not
reporting them on that basis alone.
> This is even more complicated when you want to compare two groups. For
> example, we want to analyze age related effects in performing two tasks A
> and B. We want to see regions activated by B in young group but not in old
> group. In the literature, I see that people use a contrast like [-1 1 1 -1]
> for this comparison. But there is a danger of finding common regions
> activated by task B in youngs and by A in old group. RFX might be a
> solution to this but still I'm not comfortable.
I think this is a slightly different question. The contrast you are
interested in is a task-by-group interaction. You would like to find areas
that show an interaction due to a change in brain activity in task B, with
task A constant across groups. But the concern is that the interaction may
actually isolate regions where B>A in the young group and A>B in the old
group i.e. a crossover interaction. Again, this issue is not unique to the
study design you suggest but a more general question that I have had many
long discussions about, but no simple solutions I'm afraid!
Dr. Geraint Rees
Wellcome Advanced Fellow,
California Institute of Technology,
voice (626) 396-2880
fax (626) 796-8876