[Eeglablist] Unfold: A new toolbox for overlap correction, (non)linear modeling, and regression-based EEG analyses

Benedikt Ehinger behinger at uos.de
Tue Nov 5 08:21:56 PST 2019

Dear EEGLAB users,

we are pleased to announce the official release of "unfold", a new
open-source and EEGLAB-compatible toolbox for regression-based EEG analysis.

Main features:
-- Deconvolve overlapping potentials (e.g. from stimulus onsets, button
presses, microsaccades)
-- Easy model specification with Wilkinson formulas: "y ~ 1 +
cat(stim_type) + spl(stim_contrast,5)"
-- Model influences of continuous covariates ("regression-ERPs", e.g.
Pernet et al., 2011)
-- Model non-linear predictors with spline regression (GAM)
-- Can be applied to raw EEG or to ICA activations
-- Advanced options for regularization, temporal basis functions,
circular splines
-- Documentation & tutorials available

Why is this useful?
The EEG is increasingly recorded in complex, mobile, and
quasi-experimental situations. Examples are free viewing experiments
with combined eye-tracking, experiments in virtual reality, or
brain/body imaging studies.

The resulting datasets are typically confounded by (1) strong temporal
overlap between the brain responses evoked by different stimuli and
motor actions and (2) the influences of various linear or non-linear
covariates (e.g. saccade size) on the measured neural responses.
However, even "traditional", highly-controlled EEG datasets often
contain overlapping potentials from several processes such as stimulus
onsets, microsaccades, and button presses.

While ICA can disentangle spatially overlapping neural processes, it
does not solve the problem of temporal overlap, which often differs
between conditions. Fortunately, regression-based approaches (e.g. Burns
et al., 2013; Smith & Kutas, 2015a,b; see also the EEG LIMO toolbox)
provide a flexible framework to deal with this problem.

The "unfold" toolbox deconvolves overlapping potentials and controls for
the influences of both linear and non-linear covariates on the EEG. It
directly reads EEGLAB datasets and the regression model can then be
intuitively specified with Wilkinson-formulas.

The toolbox can be downloaded here: http://www.unfoldtoolbox.org
Tutorials are found here: http://www.unfoldtoolbox.org/toolboxtutorials.html
The code is maintained here: http://github.com/unfoldtoolbox/unfold

The reference paper was just published here:
Ehinger & Dimigen (2019) Unfold: An integrated toolbox for overlap
correction, non-linear modeling, and regression-based EEG analysis.
PeerJ, http://peerj.com/articles/7838

Three example applications to combined eye-tracking/EEG data can be
found here:

We hope you find this tool useful and appreciate any feedback and bug

Benedikt Ehinger & Olaf Dimigen
The Donders Institute for Brain, Cognition and Behaviour, Nijmegen /
Osnabrück University / Humboldt-University Berlin   

More information about the eeglablist mailing list