Trial rejection

Removing EEG-artifact-laden trials from the data before analyses is an important step in preprocessing. In this lecturelet, you'll see some examples of artifacts in EEG data.
For more online courses about programming, data analysis, linear algebra, and statistics, see sincxpress.com/

[@lizcruz2909]

Thank you! This is exactly what I was looking for.

[@mikexcohen1]

Awesome :)

[@jabitte1]

thank you so much, this really helps! i love your calm and precise way of explaining things

[@mikexcohen1]

:)

[@jingyang4788]

Thanks for your detailed explanation, I found that when I remove trials, I always have different standards. My standards will be strict at first, but when I saw too many trials being deleted, I will loosen standards.

[@mikexcohen1]

Hi Jing. Yeah, that's the main downside of manual trial rejection. My recommendation is not to go through too many datasets at once. For example, if you have 20 datasets to clean and each one takes 20 minutes, better to do two per day than to do all 20 in one marathon session.

[@MAYMOUNAHBANAT]

Thanks for your deep explanation. Can I do the trials' manual rejection after already doing ICA artifact rejection, or this shall be done before?

[@mikexcohen1]

Ideally before, because otherwise you're exposing ICA to unclean data. But after is also fine.

[@MAYMOUNAHBANAT]

Thanks a lot for your continuous support. I have another question, if I have a channel of interest with about 35% bad segments, and need to run ICA, shall I exclude it from ICA then interpolate it? and what is the estimated bad segments percentage in a channel that I shall exclude from ICA since these bad segments seems to be a channel noise due to misconnection@@mikexcohen1

[@mikexcohen1]

That's a tough decision. If you think the data contain real data plus noise, then try keeping it in the ICA and check whether ICA can isolate a component just for the noise (without removing too much real signal). If you don't find a component, or if the channel is mostly noise with very little brain signal, then it's best just to remove/interpolate it.

[@MAYMOUNAHBANAT]

This helps a lot. great thanks@@mikexcohen1