[Eeglablist] Per-trial channel interpolation and ICA?

[Tarik S Bel-Bahar]

​Hi Eric,

Interpolating before ICA indeed will feed it some "non-original" data. Thus
the general routine from the ICA-EEG camp is to not interpolate before ICA,
neither across the file or within single trials.

Note also some researchers reref before ICA, and some after. My
understanding is that the rank issue is connected to the reref step too.

Your intuition is probably correct that rank changes within single trials
due to interpolating bad channels within single trials, but let's wait to
hear from some experts.
Presumably one would need to correct for adjusted rank on a trial by trial
basis perhaps?

Have you considered running ICA first with only worst channels and worst
periods removed, and then running ICA, and then doing doing ICA cleaning,
and then attempting individual trial bad-channeling ?

Have you reviewed how your results actually differ when doing ICA on files
that have had bad channels interpolated within single trials before ICA ?

Also, I think the per-trial bad-channelling procedure was developed before
ICA was a dominant force for EEG analyses. It certainly is a way to "keep
more original data" rather than throwing out channels from across the whole
record just because they are bad in some locations.

You may have noticed a recent plugin for eeglab (maybe it came from you?)
that does single-trial bad channel detection and/or interpolation?

Looking forward to hearing about how you resolve this issue. It will
certainly help other users on the list to know how you get on.

Cheers!









​

On Mon, Mar 19, 2018 at 8:15 PM, Eric Rawls <elrawls at email.uark.edu> wrote:

> Hello EEGLAB-list,
>
> I understand that ICA can only be run on the number of channels present
> (and not interpolated; i.e. present as "actual" data) in a set of EEG data.
>
> This makes sense, because removing channels reduces the rank of the data.
> However, what effect does interpolating channels which are bad in single
> trials have on data rank?
>
> I've become curious about this because various methods (FASTER, EGI
> proprietary software, others) offer this functionality, and I'm not sure if
> it is feasible to move forward from a dataset that has been processed this
> way to an IC-based analysis.
>
> So, if a channel is interpolated on an as-needed basis rather than being
> "bad" the entire recording, how does this influence the rank of the EEG?
> What if half of the channels of a  dataset were interpolated, but in
> separate epochs?
>
> Can we move forward from this sort of artifacting schema to an analysis in
> IC-space, or does the manner of data processing preclude "knowing" the rank
> of the data and therefore an accurate IC decomposition?
>
> Thanks in advance for discussion,
> Eric Rawls, M.S.
> Graduate Research Assistant
> Department of Psychological Sciences
> University of Arkansas
>
> _______________________________________________
> 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
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20180321/94f7a132/attachment.html>