[Eeglablist] Question about preprocessing EOG channels for ICA

[Makoto Miyakoshi]

Hi Ayaka,

If you find that your suggested approach works better than the standard
approach, go ahead and use it for your publication. Why not?

However, in doing so, I recommend that you show three results: (1) No
EOG-IC removal, (2) Standard EOG-IC removal, and (3) Customized EOG-IC
removal in your paper, either in the main text or in the Supplement. Based
on the comparison, try to convince your reviewers and readers including
experts. This is the only reader-friendly way to use a novel and/or
esoteric approach, and it's better than defending your approach just by
citing a reference paper. It is general advice that when you are in doubt
about whether you should choose A or B in this kind of situation, always
choose both and show the comparison, then make a choice with justification.

It's for developers, but personally, I think the following approach makes a
lot of sense.
https://urldefense.com/v3/__https://www.sciencedirect.com/science/article/abs/pii/S0165027006002834__;!!Mih3wA!CGYdTKf_6977-YZtILPE0GUP_YM4M5OjxoGCnB8OKAZ6PB-cclykRcNtoh4ayIzEpsoLCg15JMLdo2Z6hFHuwZ1CXlY$ 
However, the current status of their out-of-the-box application and its
performance compared with what's available today is unknown.

ICA results above 13 Hz start to show correlations among ICs, and it will
get worse progressively as the frequency bins increase.
But as long as you focus on ERP components at 13 Hz and below, the standard
use of ICA, including your suggested version, is fine.

Makoto

On Sat, Jul 20, 2024 at 6:28 PM Ayaka Hachisuka via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:

> Hello,
>
> I'm wondering what your thoughts are on "aggressively filtering" only the
> EOG channels for ICA? I read this recommendation in the EEGLAB wiki (
>
> https://urldefense.com/v3/__https://eeglab.org/tutorials/06_RejectArtifacts/RunICA.html__;!!Mih3wA!HQISzYIyD-J8Zuk4m-reRqRizEexcqAumaHZbBLFjykj2A3RBjKATuJGJwNDEJi8oDDuH1q_zJsW5OJ9XG6cDw$
> , see below) and
> to save myself a step, I implemented a 1Hz high-pass filter for EOG
> channels only. The EEG channels are still filtered at 0.05Hz, my original
> parameter.
>
> It seems to work really well for detecting eye movement artifacts, and my
> data visually looks better than before after ICA, but I wasn't sure if this
> was a reasonable approach.
>
> Thanks!
>
> ----------from the EEGLAB wiki page ----------
> How to deal with the aggressive high-pass filter applied before running ICA
>
> ICA decompositions are notably higher quality (less ambiguous components)
> when the data is high-pass filtered above 1 Hz or sometimes even 2 Hz.
> High-pass filtering is the easiest solution to fix bad quality ICA
> decompositions. However, for processing EEG data (such as ERP analysis),
> high-pass filtering at 2 Hz might not be optimal as it might remove
> essential data features. In this case, we believe an optimal strategy is
> to:
>
>    1. Start with an unfiltered (or minimally filtered) dataset (dataset 1)
>    2. Filter the data at 1Hz or 2Hz to obtain dataset 2
>    3. Run ICA on dataset 2
>    4. Apply the resulting ICA weights to dataset 1. To copy ICA weights and
>    sphere information from dataset 1 to 2: First, call the Edit → Dataset
>    info menu item for dataset 1. Then enter *ALLEEG(2).icaweights* in the
> *ICA
>    weight array …* edit box, *ALLEEG(2).icasphere* in the *ICA sphere array
>    …* edit box, and press *Ok*.
>
> ICA components can be considered as spatial filters, and it is perfectly
> valid to use these spatial filters on the original unfiltered data. The
> only limitation is that since strong artifacts affect low-frequency bands
> filtered out before using ICA, they may not be removed by ICA. In practice,
> we have never found this to be a problem because artifactual processes that
> contaminate the data below 2 Hz also tend to contaminate the data above 2
> Hz.
> _______________________________________________
> 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