I would like to hear some thoughts from the ResearchGate community about dealing with ICA-based artifact rejection in recordings from multiple days. As an example, I have data from one experiment where subjects was tested over 6 different days.
The choice would be whether to estimate ICA filters on each separate day or merging a subset of trials from the whole set of recordings. Curious to hear comments and possibly some reference to look at!
Thanks!
Dear Clara,
this is not an easy question to answer. In my opinion it depends on the nature of the different recordings: did you apply the same paradigm at the same time of the day, then probably you will get good results if you just merge the recordings into one big file and do ICA. If there is a difference in the recordings, e.g. open eyes, closed eyes, different paradigms, then it is better from my experience to apply ICA for every single recording.
It also is important to know for what kind of artefacts you are looking. Is it "just" the eye-movements you want to get rid of? Or do you want to exclude components from e.g. noisy channels? In the latter case there might be a problem with one merged big file, since not during every recording there are the same technical artefacts while in the case of the eye-movement artefacts there might be a good overlap for all recordings (if done during the same time of the day...).
Let me know what your thoughts are on this topic...
Sebastian
Thanks Sebastian!
I am in agreement with what you are saying. I usually make sure to merge only recordings of 'similar paradigms'. For instance, I would run different ICAs on motor tasks, visual tasks or resting state. In this case, I have the same paradigm performed on different days (a day1/day2 design repeated for three different sessions, one week apart). Right now I am going with a single merge for at least the two consecutive days. Regarding the artifacts, I am mainly interested in the ocular and myogenic ones.
My concerns are: if I run separate ICAs and then I make slightly different choices in the ICs to reject, I might 'bias' the signal on one day and not the other. So that is why I felt better going with this choice. However, ICA in general is a useful but tricky business and I guess there is no 'perfect' way to go ;) - I enjoyed reading your comment on Neuroimage about myogenic artifacts.
Thanks for your contribution and curious to hear if anyone else has something to share.
Clara
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS