I am trying to localize sources of brain activity using MEG data. I first want to compute ICA and then localize independent components of interest.
However, I sometimes have trouble making a decision about what to do.
First, I was wondering whether to perform a dimension reduction before ICA (using PCA). I tried with and without and the results are clearly different but I don't know which to "trust". I have 245 out of 248 channels that are not artifacted so after ICA computation without PCA, I am left with 245 independent components. The thing is I have many frontal components that correspond to eye movements (according to their topography and rare and abrupt increases in amplitude) : components 1,2,3, 6, 7, 11... and more. Their topography look almost exactly the same. I don't know how to interpret it. Even when I perform PCA (100 components) before ICA, I am left with several eye movement artefacts but not that many.
Could it be a problem with the raw data, due to movements during the experiment ? The machine said the subject barely moved though...
Given that I am using ICA for an other purpose than artifact removal, is it actually a problem to have so many frontal artifacts?