[Eeglablist] ICs removal

Tarik S Bel-Bahar tarikbelbahar at gmail.com
Wed Oct 23 11:47:22 PDT 2019

Hi Paolo...

Some quick suggestions re your question....
1) if you can't remove the noise otherwise at collection or via a cleaning
and 2) if you have isolated the noise into mainly one IC,
and 3) if the component is mainly noise
and 4) if you have major clear alpha-dominated components that cover more
variance and have good neural-IC properties..
and 5) if the spectral component does not vary with your experimental
conditions or groups of interest
...THEN it is probably safe to remove it, as long as you use the same rules
for other subjects/files.

also, if you haven't, try your analysis with and without clean_rawdata
plugin/function before ICA, and see what happens to the noise and the

Extra thoughts...
As one often finds, ICs are mixed, as ICA is not a perfect/pure carver of
spatial regularities in the surface potentials.
Some groups keep all ICs except ones that are clearly artifactual. Some
groups remove all ICs except those that are clearly neural.
Some groups take a more cautious approach and keep mixed ICs, or as many as
possible, to "preserve" potential loss of (neural) information. not many
published comparisons exist.
Please look into reliability procedures for ICA (e.g., relica, eegift), and
comparisons of ICA and PCA approaches.
Overall, one good strategy to try would be to remove any ICs that IClabel
plugin says is under 90% on the neural classifier, and/or over ~15% on any
artifact classifier.
One can run SASICA for complementary ICA feature diagnostics.
If you are beginning with ICA, one good strategy is to just follow the best
2 or 3 procedures from high-quality research groups that use ICA for
purposes similar to yours (e.g., using our friend Google Scholar).
if you haven't had a chance to check out the IClabel classification site
which has really good examples, tutorials, and training regarding ICs.

A useful resource to remember that helps me sometimes is the  googlable
eeglablist archives, where one can find many interesting past thread
related to one's topic of choice.

On Tue, Oct 22, 2019 at 4:18 AM Paolo Antonino Grasso <
paolo.grasso at iusspavia.it> wrote:

> Hi all,
> I have a question about IC removal
> In some cases I have found situations like the one described in the
> attached figure where the signal is very noisy (although a PSD analysis do
> not reveal the presence of electrical noise around 50 Hz) and ICA finds a
> component that, if removed, eliminates most of the noise. However, this
> component seems to retain some neural signal (as evident from the peak
> around 12 Hz and from the ERP) and I don't know if it is appropriate to
> remove it.
> Do you have any idea of what is the origin of this component/noise and how
> should I deal with it?
> Thanks a lot for your replies
> Paolo
> --
> *Paolo A. Grasso, Ph.D.*
> *Postdoctoral Research Fellow*
> *Scuola Universitaria Superiore IUSS Pavia*
> Palazzo del Broletto
> Piazza della Vittoria,15 - 27100 Pavia
> www.iusspavia.it
> Personal Web Page
> <
> http://www.iusspavia.it/people-research-fellows/-/asset_publisher/GFd4VCR7Erp7/content/paolo-a-grasso?redirect=%2Fpeople-research-fellows
> >
> skype: paolograsso86
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu

More information about the eeglablist mailing list