Dear FSL experts,
I am using concat ICA to analyse task-based fMRI data. Since the principle is looking for common spatial pattern when time series are different (different stimuli), I use dual-regression after concat ICA. However, concat ICA require all functional data must be in MNI space. I want to retain as many as information in subject native space, so I am trying to do as follow:
-Run single ICA for each subject.
-Register all the IC map of all subject into MNI space.
-Calculate and try to find which IC of this subject is highly correlate with which IC from other subject (for example if IC1 of subject 1 show strong activation in the lateral parietal lobe as well as the IC9 of subject 2 and IC5 of subject 3 do, I will group all the 3 IC in one group).
-Get time series of the IC map above from single ICA result (since the single ICA is run in subject space, I expect the information is better than when is registered into MNI space like in concat ICA). Since the task is a game like shooting spaceship, the stimuli is a huge mess and I want to get the time series (of fMRI data) first before trying to compare to different way of manipulate the stimuli. That's why I use ICA instead of GLM.
-Compare the time series with some of task stimuli.
I am stuck at step 3, finding which IC is highly correlated (or look alike) with which. May you share me the way to do that. I am thinking of calculate mutual information but have no idea how to do it.
Best regards,
Khoi, Huynh.
|