Hi
There are two possibilities that I can see:
(i) concatenate subject-wise (runA runB), making sure that A always corresponds to trial order 1 etc) and then run TICA. The problem is that data concatenation is almost always a bad idea, for various reasons discussed many times on the list (e.g. interaction with filter etc) but it's worth a try IMHO
(ii) simply run concatICA, then, take the files tXX.txt - these contain the temporal responses in the final 2N columns (N=#subjects), the first column being the eigenvector of these 2N temporal responses. You can test these files against a design matrix (e.g. created with GLM) quite easily using fsl_glm:
fsl_glm -i tXX.txt -d design1.mat -c design1.con -o test --out_z=test_z
where design1 is the design that corresponds to trial order 1. test_z then contains z_stats for each column in tXX.txt and each contrast specified in design1.con. Do the same with design2, collect all the output files and run 2sample ttest across these.
hth
Christian
On 6 Jul 2010, at 16:26, Allison Jack wrote:
> I'd just like to chime in that I also would be very interested to hear any insight anyone has on Michael's issue re: specifying the timeseries for an ICA analysis when trials are randomized across runs but not subjects (i.e. order of trials in run 1 is different from order in run 2, but run 1 and run 2 are the same for each subject)--I am struggling with the exact same thing at the moment. I have considered going in and manually concatenating the 3 column format timing files for each subject as well as their 4D functional data, so that everything is in order but in one file, but I'm not sure this is the right way of solving this problem. ~ Allison
|