[Eeglablist] Rejecting Artifacts

Tarik S Bel-Bahar tarikbelbahar at gmail.com
Sun Oct 8 11:43:38 PDT 2017


Hello Konstantina, various notes below, please take a good look. After you
fully review and try the following, please feel free to get back with
new/additional questions.


**********NOTES FOR KONSTANTINA*************************

Overall, you should take a minimum of 10 to 20 hours learning/practicing
how to do this, and most likely minimum 5 to 10 hours minimum for any
additional steps.. I usually have students practice for at least 50 hours
of cleaning with iterative feedback, etc.. before letting them truly clean
data for analyses purposes. Anything worth doing is worth doing well :)
Also, try not trust the engineering tendency to think that human biosignals
are just like other data. They are unique, have a lot of variability, and
should not be processed automatically (especially when one is a newbie).

If you are having trouble finding the right resources, make sure you are
looking at the current eeglab tutorials, have read through/gone through
most steps, and have used the eeglab tutorial data (single subject and
study sets) for practicing.

0. Overall, rejecting large multi-channel artifacts before ICA is easy.
Just find the worst periods (as per online eeglab tutorials), select them
as you are scrolling through the data, reject them, and run ICA. If you
miss some periods, it will be clear which ones you missed when you look at
the single trial summaries for different ICs.


1. The googlable eeglab tutorial on rejecting artifacts is easy to read and
understand, and if you follow the steps and instructions you should be
fine. You should practice with eeglab tutorial data if you have not had a
chance to yet.

2. It totally depends on what kind of artifacts you are looking for and
what you will use the processed data for (e.g., channel vs. ICA level
analyses).

3. If you properly clean the data of major artifacts before running ICA
(see online tutorial for that) then you will be able to rejects artifactual
ICs and that will make it easier to clean your data.

4. If you haven't had a chance to yet, learn/review/practice all the
materials on the ICA classification training site. Especially look at the
examples. I recommend to students to do at least 500 classifications. Make
use of the "practice mode with feedback" and the extensive tutorial on what
are "good" and "bad" ICs.
The site is at: reaching.ucsd.edu:8000/tutorial

5. There are several ICA classification plugins that can help you
semi-automatically detect bad ICs to reject, including ADJUST and SASICA.
Please review the articles on each that you can find easily on Google
Scholar, and do some practice with them when you have done ICA on your data.

6. Remember ICA cleaning can take care of major artifact classes including
blinks, lateral eye movements, and EMG/muscle artifacts. Do note that it's
important to understand that ICA is not perfect, and some ICs will be
mixed, but that's a higher level topic for after you have advanced your
knowledge.

7. The reject continuous data option in the eeglab GUI (using the frequency
parameters) works quite well for finding the biggest/worst periods, which
is the main/only cleaning really required before ICA. That option is under
Tools > Automatic continuous rejections. You will have to play around with
it a bit, it's critical to not use any tools blindly. Trust but verify :)

See the eeglab online summer school videos unspecific topics if you haven't
had a chance to yet.
See Makoto's pipeline suggests too if you haven't for some general hints
for newbies.
See the ASR method which is an eeglab plugin, which one can try to use for
fully automatic cleaning (but it has caveats like everything else). It's
part of the "clean_raw_data plugin" and the "PREP plugin".
Review the extensive ERPlab materials and introductions.
Read EEG handbooks from Luck, from Mike Cohen, from Handy, which each have
"chapters for newbies" but also cover advanced topics.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20171008/043f61b1/attachment-0001.html>


More information about the eeglablist mailing list