Hello,
I am using SPM5 to analyze some fMRI data. I have 9 subjects who I image
listening to different types of speech at two different moments ( before and
after training for one of the speech types).
At the first level analysis I have created a model for each subject and defined
4 contrasts of interest and saved the con_000* files for each of the 4 contrast
for each of the 9 subjects. These are the ones I am taking to the second level.
At the second level I am interested in two things mainly: (1) Differences
between sessions for 2 of the contrasts, and (2) Similitudes (areas in common)
between sessions for the 2 other contrasts. Here is what I am trying to do:
(1) LOOKING FOR DIFFERENCES: I have done a 2 paired-t-tests analysis ( one per
contrast of interest), with nine pairs each. Each pair is formed by the two
contrasts of each subject ( before and after training). Then using the contrast
[0 0 0 0 0 0 0 0 0 1 -1] and [0 0 0 0 0 0 0 0 0 -1 1] I get the areas
significantly more active on session 1 and on session 3 respectively. Does this
sound correct to you?
(2) LOOKING FOR COMMONALITIES: My main concern is at this point of the
analysis. These are the options I have though about and the problems I find:
(2.a) If I do independent one-sample t-test for each contrast of interest, then
I cannot do conjunction analysis at the second level, becuase the contrasts I
want to join are saved in different SPM.mat files.
(2.b) If I use paired t-test, I have no idea of which contrast would help me
find similarities instead of differences, as the contrast manager does not allow
me to define a contrast such as [0 0 0 0 0 0 0 0 0 1 1].
(2.c) If I use a two sample t-test, with group 1 being the contrasts of
interest in the pre-training session for the nine subjects and group 2 the same
contrasts for the nine subjects at the post-training session. Then using a
contrast [1 1] I think I am getting the areas that are significantly active for
both sessions, but I think this is wrong because this test assumes the groups
have different subjects and in my case they aren't.
Could please somebody tell me how I can better approach this type of analysis?
Thanks very much in advance for your time and help,
Javier Gonzalez Castillo
PhD Student - Biomedical Engineering - Purdue University
|