[Eeglablist] Announcing GEDAI: A new EEGLAB Plugin for unsupervised artifact correction

Ros, Tomas dr.t.ros at gmail.com
Mon Nov 10 08:35:28 PST 2025 

Dear Jinhan,

Thank you for your positive feedback on GEDAI, it's much appreciated! As
well as for sharing the challenges you are facing, which is also the kind
of feedback we were hoping for. Our goal is to iteratively improve the
GEDAI plugin so that it can best serve the EEG community.

Please see below the answers to your specific questions (which we have
highlighted in blue):

*1. Interpretation of SENSAI Scores and Noise Quantification*
I understand that SENSAI reflects the proportion of clean EEG data retained
after GEDAI denoising. Is there an additional metric or method to quantify
the extent of noise removed? For example, ICLabel provides counts of
independent components removed due to specific artifact types (e.g., eye
movements, muscle activity, line noise). Similarly, I’m interested in
obtaining objective measures that compare noise levels before and after
GEDAI processing.

Yes, good point. Although SENSAI may be a good indicator of relative
denoising performance within an EEG recording, it is not the best option
for comparing between recordings (given the different "absolute" fit
between a specific recording/subject and the theoretical model).  GEDAI
does not currently categorize the rejected components as of a particular
type, but simply discards them if they are considered as noisy (i.e. not
brain-generated). This is done both at an epoch-by-epoch level (e.g. every
second) as well as across multiple wavelet bands (e.g. 6-10 bands depending
on sampling rate). Hence, there are quite a lot of components to account
for! One idea could be to run (ICA + IClabel) separately on the original as
well as the denoised data, and compare if there is an increase in the
relative proportion of "brain" components.
Finally, according to your suggestion, here is a potential objective
measure of noise removal:

*Explained Noise Variance (ENOVA)*
This is simply the variance of the removed noise, expressed as a proportion
of the variance of the original EEG data (higher values indicate more noisy
original data).
*Matlab Code:*    [EEGclean, EEGartifacts ] = GEDAI (EEGoriginal); ENOVA =
var( EEGartifacts.data(:) ) / var( EEGoriginal.data(:) )

*2. Delta Band Power Removal and Aperiodic Signal Modeling*
Our group focuses on modeling periodic and aperiodic (1/f) components of
resting-state EEG using the FOOOF algorithm. When applying GEDAI to our
resting EEG datasets, we noticed substantial attenuation of delta band
power (1–4 Hz). This reduction appears to affect the accuracy of aperiodic
modeling, as the cleaned power spectral density no longer exhibits a clear
1/f trend across the 1–40 Hz range. Could you clarify why GEDAI treats
delta band activity as noise, and whether there are options to preserve
low-frequency components more selectively?

Admittedly, this first version of the GEDAI plugin removes a significant
part of the lowest wavelet frequencies (resulting in the removal of delta
activity <2  Hz). This was a compromise related to the fixed 1-second epoch
length across bands, since there is a tradeoff between estimating the
covariance at shorter (e.g. gamma, benefitting from smaller epoch sizes)
and longer (e.g. delta, benefitting from longer epoch sizes) timescales. As
a result, we have now updated the plugin to GEDAI v1.2
<https://urldefense.com/v3/__https://github.com/neurotuning/GEDAI-master/releases/tag/v1.2__;!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMg_8Ca07g$ > (download
at https://urldefense.com/v3/__https://github.com/neurotuning/GEDAI-master/releases/tag/v1.2__;!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMg_8Ca07g$ )

This includes the following changes:

1. No more cropping of data if EEG length is not a whole number of epochs
(incomplete epochs are denoised using reflective padding)
2. Epoch size is not fixed any more across wavelet bands (it is based on
the number of wave cycles of the wavelet-band center frequency)
3. The low-cut/high-pass frequency can now be explicitly defined by the user

Modified inputs to GEDAI.m :
epoch_size_in_cycles - Epoch size in number of wave cycles for each wavelet
band. Default is 12.
lowcut_frequency - Low-cut frequency in Hz. Wavelet bands below this
frequency will be excluded. Default is 0.5 Hz.

*GEDAI does NOT need additionally to use other filters, such as IClabel,
ASR, Automatic Epoch Rejection?*
Well, this is of course a more complicated question, that will require
wider testing by the community. Although the initial findings from the preprint
<https://urldefense.com/v3/__https://www.biorxiv.org/content/10.1101/2025.10.04.680449v1__;!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMiubZILaQ$ >indicate that
GEDAI can be used in *isolation *(and without explicit bad channel
rejection), we have not analysed how it may be used in *combination *with
pre-existing tools to potentially further improve denoising performance.
This is still very much an open question and the customary phrase applies
that further research is needed .

Best wishes and please let us know how it goes,

the GEDAI development team (Tomas Ros, Victor Férat, Yingqi Huang, Abele
Michela)

▬▬▬


*Tomas Ros, PhD*

Lecturer, Department of Clinical Neurosciences
Research Staff Scientist, CIBM EEG HUG-UNIGE Section
University of Geneva, Switzerland
https://urldefense.com/v3/__https://www.tomasros.com/__;!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMgGuNSDbA$ 

[image: CIBM logo]


*Clinical and Translational Neuroimaging*

UNIGE Campus Biotech

9 Chemin des Mines

CH - 1202 Genève, Switzerland


(+41) 0766 158 863



cibm.ch <https://urldefense.com/v3/__http://www.cibm.ch/__;!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMjyuxcnrQ$ >

*Follow us:  *[image: Twitter] <https://urldefense.com/v3/__https://twitter.com/CIBM_ch__;!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMjgbEP-iQ$ >  [image:
LinkedIn]
<https://urldefense.com/v3/__https://www.linkedin.com/company/cibm-center-for-biomedical-imaging/__;!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMjfy2m3DA$ >  [image:
YouTube] <https://urldefense.com/v3/__https://www.youtube.com/channel/UCA8rMvimtkVzFwWzDDlaGMQ__;!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMj9PnKHqg$ >






On Wed, 5 Nov 2025 at 20:54, Park, Jinhan <jinhan.park at ufl.edu> wrote:

> You don't often get email from jinhan.park at ufl.edu. Learn why this is
> important <https://urldefense.com/v3/__https://aka.ms/LearnAboutSenderIdentification__;!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMhhB2Md6g$ >
> Dear GEDAI Development Team,
> I am a postdoctoral researcher at the University of Florida, and I would
> like to express my appreciation for your development of the GEDAI plugin.
> It has proven to be a fast and effective tool for cleaning EEG data, and
> our research group is seriously considering integrating GEDAI into our
> current EEG preprocessing pipeline.
> However, I’ve encountered some challenges in interpreting the SENSAI
> scores and understanding certain aspects of the denoised data—particularly
> regarding delta band power. I would be grateful for your insights on the
> following questions:
>
>    1. *Interpretation of SENSAI Scores and Noise Quantification*
>    I understand that SENSAI reflects the proportion of clean EEG data
>    retained after GEDAI denoising. Is there an additional metric or method to
>    quantify the extent of noise removed? For example, ICLabel provides counts
>    of independent components removed due to specific artifact types (e.g., eye
>    movements, muscle activity, line noise). Similarly, I’m interested in
>    obtaining objective measures that compare noise levels before and after
>    GEDAI processing.
>    2. *Delta Band Power Removal and Aperiodic Signal Modeling*
>    Our group focuses on modeling periodic and aperiodic (1/f) components
>    of resting-state EEG using the FOOOF algorithm. When applying GEDAI to our
>    resting EEG datasets, we noticed substantial attenuation of delta band
>    power (1–4 Hz). This reduction appears to affect the accuracy of aperiodic
>    modeling, as the cleaned power spectral density no longer exhibits a clear
>    1/f trend across the 1–40 Hz range. Could you clarify why GEDAI treats
>    delta band activity as noise, and whether there are options to preserve
>    low-frequency components more selectively?
>    3. *GEDAI does NOT need additionally to use other filters, such as
>    IClabel, ASR, Automatic Epoch Rejection? *
>
> Thank you very much for your time and for developing such a promising
> tool. I look forward to your response.
> Best regards,
> Jinhan Park
>
>
>
> *Jinhan Park (Hans) | Ph.D. *
>
> Department of Applied Physiology and Kinesiology
>
> University of Florida
>
> Laboratory of Rehabilitation Neuroscience | lrnlab.org
>
> Phone: +1 (352) 328-6605
>
> Email: jinhan.park at ufl.edu
>
>
> ------------------------------
> *From:* eeglablist <eeglablist-bounces at sccn.ucsd.edu> on behalf of Tomas
> Ros via eeglablist <eeglablist at sccn.ucsd.edu>
> *Sent:* Monday, October 20, 2025 2:53 PM
> *To:* eeglablist at sccn.ucsd.edu <eeglablist at sccn.ucsd.edu>
> *Subject:* [Eeglablist] Announcing GEDAI: A new EEGLAB Plugin for
> unsupervised artifact correction
>
> [External Email]
>
> Dear EEGLAB community,
> We are writing to announce our new plugin, the Generalized Eigenvalue
> De-Artifacting Instrument (GEDAI), which is now available in the Extension
> Manager.
> GEDAI was developed to provide robust and rapid artifact removal, even for
> highly contaminated data, without requiring user supervision. It introduces
> a novel approach called "leadfield filtering," which uses a theoretical EEG
> forward model to separate brain activity from artifacts.
> GEDAI was designed to address key limitations of current ICA- or PCA-
> based denoising, such as speed, challenging artifact mixtures, or need for
> calibration data.
> Our benchmark tests<
> https://urldefense.com/v3/__https://www.biorxiv.org/content/10.1101/2025.10.04.680449v1__;!!Mih3wA!AUK0DyQeIgIlJl7UGbrzSjRIorBGG-rXomeQ_Z4PbNBQiWBRFkCrHDO-8gYgLImNW83Yh6iO78ayvq0yjDaFEJxo2w$
> > using real-world data indicate that GEDAI preprocessing can result in
> more accurate neurobehavioral prediction compared to competing methods.
> Nevertheless, it has not been tested under all EEG scenarios and we would
> welcome both positive and negative feedback on its performance.
> You can also download the plugin and find its documentation here:
> https://urldefense.com/v3/__https://neurotuning.github.io/GEDAI-master/__;!!Mih3wA!AUK0DyQeIgIlJl7UGbrzSjRIorBGG-rXomeQ_Z4PbNBQiWBRFkCrHDO-8gYgLImNW83Yh6iO78ayvq0yjDYaj2FnGg$
> <
> https://urldefense.com/v3/__https://github.com/neurotuning/GEDAI-master__;!!Mih3wA!AUK0DyQeIgIlJl7UGbrzSjRIorBGG-rXomeQ_Z4PbNBQiWBRFkCrHDO-8gYgLImNW83Yh6iO78ayvq0yjDYf-A47VA$
> >
> We hope GEDAI becomes a valuable tool for the community.
> Best regards,
> the GEDAI development team
>
> ▬▬▬
>
>
> Tomas Ros, PhD
>
> Lecturer, Department of Clinical Neurosciences
>
> Research Staff Scientist, CIBM EEG HUG-UNIGE Section
> University of Geneva, Switzerland
>
> https://urldefense.com/v3/__https://www.tomasros.com/__;!!Mih3wA!AUK0DyQeIgIlJl7UGbrzSjRIorBGG-rXomeQ_Z4PbNBQiWBRFkCrHDO-8gYgLImNW83Yh6iO78ayvq0yjDZ-36KFHg$
>
> [CIBM logo]
>
> Clinical and Translational Neuroimaging
>
> UNIGE Campus Biotech
>
> 9 Chemin des Mines
>
> CH - 1202 Genève, Switzerland
>
> (+41) 0766 158 863
>
>
>
> cibm.ch<
> https://urldefense.com/v3/__http://www.cibm.ch/__;!!Mih3wA!AUK0DyQeIgIlJl7UGbrzSjRIorBGG-rXomeQ_Z4PbNBQiWBRFkCrHDO-8gYgLImNW83Yh6iO78ayvq0yjDZzuwrnAw$
> >
>
> Follow us:  [Twitter] <
> https://urldefense.com/v3/__https://twitter.com/CIBM_ch__;!!Mih3wA!AUK0DyQeIgIlJl7UGbrzSjRIorBGG-rXomeQ_Z4PbNBQiWBRFkCrHDO-8gYgLImNW83Yh6iO78ayvq0yjDbpSPRxbQ$
> >   [LinkedIn] <
> https://urldefense.com/v3/__https://www.linkedin.com/company/cibm-center-for-biomedical-imaging/__;!!Mih3wA!AUK0DyQeIgIlJl7UGbrzSjRIorBGG-rXomeQ_Z4PbNBQiWBRFkCrHDO-8gYgLImNW83Yh6iO78ayvq0yjDa2ocjhWw$
> >   [YouTube] <
> https://urldefense.com/v3/__https://www.youtube.com/channel/UCA8rMvimtkVzFwWzDDlaGMQ__;!!Mih3wA!AUK0DyQeIgIlJl7UGbrzSjRIorBGG-rXomeQ_Z4PbNBQiWBRFkCrHDO-8gYgLImNW83Yh6iO78ayvq0yjDa4Ems7Ow$
> >
>
>
>
> _______________________________________________
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu or visit
> https://urldefense.com/v3/__https://nam10.safelinks.protection.outlook.com/?url=https*3A*2F*2Fsccn.ucsd.edu*2Fmailman*2Flistinfo*2Feeglablist&data=05*7C02*7Cjinhan.park*40ufl.edu*7C26b5ea14565f42dfecb608de103d3373*7C0d4da0f84a314d76ace60a62331e1b84*7C0*7C0*7C638966052243284833*7CUnknown*7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ*3D*3D*7C0*7C*7C*7C&sdata=39*2BQOMTVUrhuc1V*2BMA70ies5O*2FDKT9o6uwD17fTkLwY*3D&reserved=0__;JSUlJSUlJSUlJSUlJSUlJSUlJSUlJSUlJSU!!Mih3wA!FBuQmdCya-rH-4mOWRHLw1U39lY5ctjYHlDVKgjZmzNWB4lExChtVNV7WczAChZFUpre1YO-uKw8OMh4OxPi_w$ 
> <https://sccn.ucsd.edu/mailman/listinfo/eeglablist >.
>

More information about the eeglablist mailing list