Dear FSL users,
I am performing ICA on resting state fMRI datasets of epilepsy patients. I do understand how the FastICA algorithm, as implemented in FSL, works. The independent components are stored in matrix S (As we assume the matrix equation of ICA to be X=AS, with X the fMRI data and A the mixing matrix). I actually do not understand how the transformation of these S-matrix into Z-statisctics works. In the technical report (Beckmann, Smith) I read that the transformation occurs by dividing the raw IC estimate by the estimate of the voxel-wise noise standard deviation. I don't understand why this makes sense and I don't understand what the Z statistic value really means. In the end, the default threshold of ICA is set to 0.5. How can this be translated to thresholding based on the chance that the voxel belongs to the (noisy) background?
I hope someone can help me out.
Thanks in advance.
Debby
|