My problems with SPM are two-fold.
Firstly, I don't understand F-contrasts well. I thought I did but now I am not
sure I do.
Secondly, I don't know how to do a GLM analysis on multiple subjects. Let me
elaborate further.
So far, what I'm doing is simple: subjects answered some questions on a
computer screen; questions were either "counting" questions (they must press
a button to show how many objects they see on the screen) or "reasoning"
questions (they must choose one of several answer choices that best fit a
given pattern); a block design was used, a question was displayed for 9
seconds (during which the subject had to press a button to answer) and there
was a 3-second rest period after each question; 36 questions in all.
From my understanding of what the SPM5 manual says, T-contrasts are
suitable when I use only the canonical HRF, and F-contrasts are suitable when
I use the canonical HRF and its partial derivatives. F-contrasts are also the
way to go for any analysis of multiple subjects.
I am also of the understanding that GLM of multiple subjects is a 2-level
process: I have to do a 1st level analysis on each subject (presumably, using
only F-contrasts), generate contrast images from each subject, and then use
those contrast images in a 2nd-level analysis.
There is one example in the manual (Chapter 30 of the SPM5 manual) where
they show how to do a 2nd-level multi-subject analysis. In that example, got
three con*.img images from each subject; one for the canonical HRF, one for
its time derivative and one for its dispersion derivative. I really wonder how
they got those images because it appears SPM doesn't let me choose either of
the derivatives on its own. [In my case, I used an F-contrast for all canonical
HRF and all its derivatives, and I got ess*.img images and not con*.img
images.]
Well, I proceeded with my ess*.img images and did the 2nd-level analysis. I
got some results but I don't know how to make sense out of them.
The reason I don't know how to make sense out of them is my understanding
of T- and F-contrasts. Let's assume I have 3 conditions: counting, reasoning
and rest. (Another question I have is whether or not to model the "rest" as a
separate condition. So far, I've been doing that. What do you think?) Well, if I
want to, say, know what brain areas are activated during counting, my T-
contrast vector is [1 0 0]. What would my F-contrast vector be? Since I have
the canonical HRF and its two derivatives, I'm assuming my F-contrast vector
would now be [1 0 0 0 0 0 0 0 0; 0 1 0 0 0 0 0 0 0; 0 0 1 0 0 0 0 0 0]. But
when I did the 2nd-level analysis, I couldn't at all enter such an F-contrast!
So, in view of all this background, my questions are:
1. How do I conduct a GLM analysis of multiple subjects? In fact, in answering
this question, you'd automatically be answering my other questions....
2. Is my understanding of F-contrasts correct?
3. In the study I have above, would you model "rest" as a 3rd condition?
Thank you very much.
|