[Eeglablist] IC components

Tarik S Bel-Bahar tarikbelbahar at gmail.com
Sat Oct 21 11:41:42 PDT 2017


Hello Konstantina, here are brief notes below with specific recommendations
about what one needs to do understand what's going on. Thanks for getting
back to the list when you find out what the issue was, so that other people
in the list can learn from your experiences. All the best


****************BEGIN FOR KONSTANTINA*********************

0. If you haven't had a chance to yet, first tkae the (important) time to
use the eeglab tutorial and eeglab tutorial data to practice cleaning, ICA,
and ERPs. Confirm that you can detect neural and artifactual ICs in the
tutorial data, and that you can easily compute/see channel-level ERPs. Also
check out the rest of tutorials, and online eeglab videos about basic
processing (from "cleaning" to "ICA" to ERPs" at least).

1. The specific automatic rejection methodyou used (and what was rejected)
matters. So please provide more info/details if you want some comments
about that. For example, did you make epochs, and then use the automatic
epoch rejection? Or did you use the "Reject Data" option and use the
various tools there to detect possibly artifactual epochs? Did you use the
clean_data method? Etc...

2. You need to provide an example of the ICs you got in order to tell
whether or not you had a good ICA decomposition.

3. You should first see if you have any ERPs after
basic/visual/semi-automatic rejection of the worst epochs.

4. Check that you baselined the epochs correctly, and then properly saved
and updated the file after baselining (to a period before stimulus onset).

5. It's highly unusual to not have ERPs if you have even-related protocol
and your data organization and cleaning were done correctly. Go back and
confirm whether you see any evidence of ERPs, and do so BEFORE dropping any
ICs. Also go back and confirm that you data looks normal when you plot it,
that you have loaded the right channels, etc....

6. The IClabel tutorial requires one learning how to classify by practicing
on a large number of ICs (I recommend at least 500 classificatioins for
beginning students, with feedback from an expert). So try to make sure
you've done that learning correctly and enough.

7. Use also ADjust or MARA or SASICA eeglab plugins to get suggestions
about which ICs might be neural, artifactual, or mixed. Don't trust a
beginner's eyes just based on the IClabel tutorial.

8. Plot you channel ERPs without any ICA-baed rejection to
confirm/disconfirm that you are getting/not-getting ERPs.

Generally speaking, more details is better. I recommend in the future you
provide a full list of your actual steps, step-by-step.
Many things are easy to miss or "do backwrds" when beginning with EEG.

Check out also Makoto's suggestions for steps on his pipeline page, and
double-check that you have tried on pass of analyses using his suggestions
too.






On Sat, Oct 21, 2017 at 2:27 AM, Konstantina Tsekoura <
tsekou at ceid.upatras.gr> wrote:

> Hello,
>
> I would like to ask a more specific question about ICA. I cleaned my data
> in a semi-automated way by using the automatic rejection in Eeglab and by
> visual inspection and then run ICA.
> All of my ICA components categorized as "Other" according this IClabel
> tutorial "http://labeling.ucsd.edu/tutorial/labels". Do i have to remove
> them? Does this mean that my data are bad and i have to throw away all the
> dataset? Also, i plotted channel Erps and erps appear flat except the F8
> channel.
> I would be grateful if you could help to understand what's going on.
>
>
> Best regards,
> Konstantina
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20171021/56dca909/attachment-0001.html>


More information about the eeglablist mailing list