[Eeglablist] Number of independent components to removed in EEG preprocessing before different analysis method

[Tarik S Bel-Bahar]

Hello, brief thoughts below regarding your question. best wishes.

The number of ICs to remove depends on how many are left after the "neural
ICs" are identified, and the biases/preferences of the researchers who are
processing the data.

Thus if you have 128 ICs, and 20 ICs are considered neural, the remaining
108 ICs could be removed.
However, remember that you should test to see what different removal
regimes do to your final data/metrics.

Some researchers only take out ICs that are clearly artifactual, and keep
the rest they are not sure about because they want to "keep as much
information as possible." and they don't want to remove information that is
not clearly artifactual.

Other researchers remove all ICs that are not clearly neural, because they
believe that the (few) neural ICs found are the main "real information" and
the rest of the ICs are not important.

Your best approach is to copy the methods of ICA-EEG articles in
high-quality research articles, rely on them to make the best possible
decision for your data/analyses. Overall, I would say cleaner is better,
unless there is a special reason to change your cleaning method for a
specific kind of analysis.

At minimum, take out ICs that are clearly classified as artifacts (not
clearly neural). You can find information about what the characteristics of
those ICs are by reviewing articles about RELICA, SASICA, ADJUST, MARA, and
other ICA classification methods (all are easy to find via Google Scholar).

If you haven't had a chance to, review the ICA labeling tutorial examples
from LUCA below, which should help you know which components you definitely
should consider to be non-neural.
https://labeling.ucsd.edu/tutorial/about








>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20181105/6c782514/attachment.html>