Greetings,<div><br></div><div>Quick questions below about use and interpretation of ICA decompositions, </div><div>from a dense eeg experiment presenting faces on each trial.</div><div>If anyone has run across problems of this sort, and/or can envision</div>
<div> a pragmatic solution path, it would be much appreciated!</div><div>**this general topic may have come up on eeglablist before,</div><div>but I had some trouble searching through the whole archive easily.</div><div>
**a sample file,examples, and more information are available upon request,</div>
<div>but the information below should be enough details to attract a solution.</div><div><br></div><div><b>***Description of issue</b></div><div>In our AMICA decompositions, we are getting several similar ICs</div><div> (e.g., P100 topography), which are associated with</div>
<div> different parts of the recording session,</div><div>with one IC being active only on the first half of trials, </div><div>and the second being active in the second set of trials.</div><div>In other words, the erpimage of each component shows only</div>
<div>activity in the first half or second half of trials.</div><div><br></div><div>It seems that "rewetting" of electrodes in the middle of the session </div><div>led to changes in electrode placement, which leads to having these duplicate ICs,</div>
<div>due to spatial stationarity assumptions in ICA algorithms.</div><div><br></div><div>A similar issue comes up when examining ICs from three </div><div>different recording sessions of one subject, with triplets of ICs showing up,</div>
<div>but again, with each IC being active for only one session's trials, but not the rest of the trials.</div><div><br></div><div><b><br></b></div><div><b>***My basic questions are: </b></div><div><b>What are the fastest and most pragmatic solutions at this point ?</b></div>
<div><br></div><div>A. Is there a method to safely (and defendably) merge the double </div><div>or triple ICs within each subject session ? </div><div><br></div><div>B. If I just use the ICs as they are, and simply cluster them within each recording session,</div>
<div>how do I best account for the fact that IC activation </div><div>(for each double/triple IC) occurs only in a half or a third of trials, </div><div>either for within subject or across-subject comparisons ?</div><div>
<br></div><div><br></div><div>All the best,</div><div>Tarik</div><div><br></div>