Hi,
This all sounds correct but you just haven't got to the final level of your analysis, assuming you have a group of subjects. I will answer for a set of different cases (two groups of subjects, one group of subjects, or a single subject) as I'm
not quite sure what you have.
If you have multiple runs per subject then you should follow the three-level analysis approach as shown here:
From what you've written it sounds like you have got the first two levels setup correctly. All that you need to do now is to do a final level (if you have a group of subjects) where you combine across the 8 basis functions and across any groups
that you have.
So in your case the final level design matrix will take, as inputs, 8 basis functions (the outputs of the 8 copes you setup in the second level analysis) for every subject. Then you would do a group average for each basis function as a separate
EV and can put an F-test across this set of contrasts. If you had two groups then you would have a separate EV per group per basis function, and then each contrast would perform the group difference for one basis function, and finally you can put an F-test
across all these contrasts. If you have one group then you just have +1 values for the appropriate basis functions in each EV, but still have one EV per basis function (averaging over all subjects) and then do the F-test across simple contrasts (one contrast
per EV, just selecting that EV with a +1 value).
If you only have a single subject then you actually don't need the third level and can just look at the F-test from the second level that you already have.
I hope this helps.
All the best,
Mark
Hi FSL experts,
I have a question on doing second-level analysis with modeling HRF with FIR. We had a block design experiment where in each run we brush the subject's hand for 30 seconds, followed by a 30s's rest, repeating the brush-rest cycle for 4 times. There were 3 runs
like this. Since the subject had stroke at somatosensory cortex, which might alter HRF, I modeled HRF with 8 FIR basis functions within a time window of 24 seconds, given 3-second's TR. Given 8 basis functions, there were 8 EVs and 8 contrasts, and a F-test
with all the contrasts turned on. For the first-level analysis, I assume I should be looking at zfstats.nii from the F-test. However, I am not sure how to get correct results when combining the 3 runs together.
What I am currently doing is feeding the 24 cope images (3 runs * 8 copes/run) into the higher-level analysis in FSL. In Full Model Setup - EVs, I set 8 EVs. The 3 cope1.nii.gz files from each of the 3 runs correspond to EV1, and the 3 cope2.nii.gz correspond
to EV2, and so on. In Contrasts and F-tests, I set 8 contrasts, each with one EV as 1. Finally, I set one F-tests, with all the contrasts turned on. I assume that zfstat1.nii.gz in the gfeat directory is what I should be looking for. But I am really not sure.
Thank you!
Yuqi