[Eeglablist] Let's test whether GEDAI is a post-ASR EEG artifact rejection champion

Wed Apr 1 18:24:17 PDT 2026

Hi Rob and Cedric,

I agree with Cedric.
Basically, GEDAI replaces ICA for data cleaning, not (weak) ASR as
preprocessing for ICA. So no need to use ICA for further artifact rejection.
Post-GEDAI ICA is of course meaningful if your goal is to extract
brain components, hence GEDAI-ICA-ICLabel is still effective as well.
Post-GEDAI ICA should work very well.

We typically use ASR as a preprocessing for ICA to address the issue of
non-stationary high-amplitude artifacts. However, if we use aggressive
thresholds, ASR can be also used for final data cleaning as well: ASR was
originally developed for online BCI, for which there is no time to make use
of 'subsequent ICA'. Of course, the quality of the process is a different
story.

I think how GEDAI works depends on how reasonable SENSAI is, without
looking into the code at all. Maybe SENSAI is one of the targets for future
customization/improvement.

MOTTO SENSAI (more delicate), SUGEE SENSAI (very delicate), and SENSAI
SUGITE KUSA (most delicate) are good names for modified SENSAI.

Makoto

On Wed, Apr 1, 2026 at 5:19 PM Cedric Cannard via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:

> Hi Rob,
>
> If you see the signal quality after GEDAI, you shouldn't see the need to
> run ASR or ICA, it corrects everything very rigorously. You'd add
> unnecessary compute time.
>
> One of the advantage I forgot to mention actually is that GEDAI is alos
> pretty fast in parallel computing mode (in my experience) compared to doing
> clean_channels() + ASR + ICA + ICLabel as in my usual pipeline. Maybe 2
> times faster, using the fast PICARD algorithm (on large recordings).
>
> A few things I wonder if someone has some answers (including the authors
> that I think are on this list)
> - should we only use GEDAI on data referenced to average (CAR), or would
> the assumptions also work for other schemes like mastoids, REST, CSD, etc.
> - can filtering be changed without disrupting the algorithms (e.g.
> highpass at 0.1, 05, 2 Hz, etc.)?
> - should we remove bad channels prior for better performance (what I did)
> or not necessary?
> - should bad channels be interpolated AFTER GEDAI to avoid potential rank
> difficient issues like with ASR and ICA, or not a problem here?
> - has it been tested on other montages with more or less electroes (e.g.
> <32, 32, 128, etc.)?
>
> Thanks in advance for any insights.
>
> Cedric Cannard
>
>
> Sent with Proton Mail secure email.
>
> On Wednesday, April 1st, 2026 at 12:10 PM, Rob Coben via eeglablist <
> eeglablist at sccn.ucsd.edu> wrote:
>
> > Very interesting. I watched the video and have reviewed to primary
> article. It certainly looks promising and the logic behind it is sound.
> >
> > If anyone has tried using this I would love to hear your thoughts and
> experiences.
> >
> > Also, any opinion on using multiple tools together. For example, using
> GEDAI and followed up with iclabel or mara. Is it possible they are
> additive or is there redundancy to make that unnecessary?
> >
> > Thanks
> >
> > Rob
> >
> > > On Mar 31, 2026, at 11:17 AM, Makoto Miyakoshi via eeglablist <
> eeglablist at sccn.ucsd.edu> wrote:
> > >
> > > Hi Cedric,
> > >
> > > Thank you for your very quick response and sharing your impression.
> > >
> > > I think one nice additional advantage of GEDAI over ASR is that it is
> > > probably more developer friendly. GEDAI is composed of more general
> > > concepts: lead field and GED, while ASR is based more on free and
> unique
> > > ideas supported by empirical parameter optimization. For example, the
> use
> > > of truncated Gaussian, geometric median across covariance matrices, or
> its
> > > Riemanian variant, etc etc., ending up with SD==20 being a de facto
> > > default, which is obviously unusual but few can explain where exactly
> it
> > > comes from.
> > >
> > > Hi Norbert,
> > >
> > > Yes, GEDAI supports EEGLAB. You can use it as an EEGLAB plugin.
> However,
> > > you have to manually download the file and move them under
> /eeglab/plugins
> > > as we used to do before the modern GUI-based management system was
> > > introduced. Please see this page. You can find both the download
> package
> > > and an instruction of how to 'install' it to EEGLAB
> > >
> > >
> https://urldefense.com/v3/__https://github.com/neurotuning/GEDAI-master__;!!Mih3wA!Ewu6iAWopqNjHfgqD4a5dp_o_Vr__gyjZ5WkzwJdtVEVKBZQjhjnGnfHoDT7gomOKfuHRXEwSD9QN5hM_p2yJKR8DKM$
> > >
> > > Makoto
> > >
> > > On Mon, Mar 30, 2026 at 6:58 PM Cedric Cannard via eeglablist <
> > > eeglablist at sccn.ucsd.edu> wrote:
> > >
> > >> Hi Makoto,
> > >>
> > >> I’ve been playing with it and was impressed. It feels like the
> thresholds
> > >> may be a little too aggressive to me (the ‘auto-‘ was better), but I
> > >> haven’t done any formal comparisons yet. I’ll share here if I do.
> I’ve been
> > >> also very curious to learn more but haven’t had the time to dig more
> yet.
> > >>
> > >>
> > >> Cedric
> > >>
> > >> Sent from Proton Mail for iOS.
> > >>
> > >> -------- Original Message --------
> > >> On Monday, 03/30/26 at 14:56 Makoto Miyakoshi via eeglablist <
> > >> eeglablist at sccn.ucsd.edu> wrote:
> > >> Hi EEGLAB mailing list subscribers,
> > >>
> > >> I watched this Youtube video on GEDAI, a relatively new EEG artifact
> > >> rejection algorithm, and got very impressed. If you are interested
> in, or
> > >> looking for, an effective artifact rejection method, you definitely
> want to
> > >> check it out.
> > >>
> > >>
> https://urldefense.com/v3/__https://www.youtube.com/watch?v=qSM5narynzc__;!!Mih3wA!CLlPjuH-5y0g4OtnG1F8KKjQoNScMgDneZVoRYZ1yoUGb_ySMhoAaf4jO0oLiDxE_IZRlBbTXuSFkSA4XgNw9TOfbZ8$
> > >> Do not miss the scene in which Tomas shows a comparison before and
> after
> > >> removing the TMS-induced artifact (about 20 min position). I've never
> seen
> > >> a more dramatic demo than this. It is only comparable to successful
> demo of
> > >> gradient artifact removal from fMRI-EEG recording.
> > >>
> > >> Here is a suggestion to this community:
> > >> If you have been already using ASR, why don't you please use GEDAI as
> well
> > >> in parallel, compare the results, and share your impression with us?
> I'll
> > >> do so in my next project and report it here.
> > >>
> > >> The reason why I like GEDAI is that the concept of combining lead
> field +
> > >> GED is elegant. The idea of 'learning from data's own cleanest part'
> is the
> > >> same as ASR, but GEDAI does it more explicitly and simply. GEDAI is
> like
> > >> ASR + REST (a reference method using a forward model).
> > >>
> > >> Makoto
> > >> _______________________________________________
> > >> To unsubscribe, send an empty email to
> > >> eeglablist-unsubscribe at sccn.ucsd.edu or visit
> > >> https://sccn.ucsd.edu/mailman/listinfo/eeglablist      .
> > >>
> > >> _______________________________________________
> > >> To unsubscribe, send an empty email to
> > >> eeglablist-unsubscribe at sccn.ucsd.edu or visit
> > >> https://sccn.ucsd.edu/mailman/listinfo/eeglablist     .
> > > _______________________________________________
> > > To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu or visit
> https://sccn.ucsd.edu/mailman/listinfo/eeglablist   .
> >
> > _______________________________________________
> > To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu or visit
> https://sccn.ucsd.edu/mailman/listinfo/eeglablist  .
> _______________________________________________
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu or visit
> https://sccn.ucsd.edu/mailman/listinfo/eeglablist .