EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis

https://doi.org/10.1016/j.jneumeth.2003.10.009Get rights and content

Abstract

We have developed a toolbox and graphic user interface, EEGLAB, running under the crossplatform MATLAB environment (The Mathworks, Inc.) for processing collections of single-trial and/or averaged EEG data of any number of channels. Available functions include EEG data, channel and event information importing, data visualization (scrolling, scalp map and dipole model plotting, plus multi-trial ERP-image plots), preprocessing (including artifact rejection, filtering, epoch selection, and averaging), independent component analysis (ICA) and time/frequency decompositions including channel and component cross-coherence supported by bootstrap statistical methods based on data resampling. EEGLAB functions are organized into three layers. Top-layer functions allow users to interact with the data through the graphic interface without needing to use MATLAB syntax. Menu options allow users to tune the behavior of EEGLAB to available memory. Middle-layer functions allow users to customize data processing using command history and interactive ‘pop’ functions. Experienced MATLAB users can use EEGLAB data structures and stand-alone signal processing functions to write custom and/or batch analysis scripts. Extensive function help and tutorial information are included. A ‘plug-in’ facility allows easy incorporation of new EEG modules into the main menu. EEGLAB is freely available (http://www.sccn.ucsd.edu/eeglab/) under the GNU public license for noncommercial use and open source development, together with sample data, user tutorial and extensive documentation.

Introduction

Though computing capabilities of nearly every electrophysiology laboratory are now sufficient to allow advanced signal processing of biophysical signals including high-density electroencephalographic (EEG) recordings, many researchers continue to rely on amplitude and latency measures of peaks in EEG trial averages, termed event related potentials (ERPs). Historically, the response averaging method was developed under technical constraints imposed by hardware initially available for psychophysiological experiments in 1950s and 1960s. Before digital computers were available, researchers had to find a way to summarize event-related activity across several EEG trials representing brain responses to sensory stimulations. For this purpose, they first used analog registers to sum activity across EEG data trials. The first computerized response averaging computer, the computer of average transients (CAT, ca. 1962) helped promote the use of response averaging, called at first sensory ‘evoked potentials’ (EPs) and later the sensory/cognitive ‘event-related potentials’ (ERPs).

Using the fast and low-cost digital computers now available, technical limitations that constrained researchers to confine their EEG data analysis to simple ERP measures and parametric statistics are no longer relevant. The rationale used to justify response averaging is that the single-trial EEG data time locked to some class of experimental events consists of an average ERP, whose time course and polarity is fixed across the trials, plus other EEG processes whose time courses are completely unaffected by the experiment events. The cortical sources of ERP features may be assumed to be spatially distinct from sources of spontaneous EEG activities. However, as we have demonstrated recently, focusing data analysis on response averages alone ignores, first, event-related dynamics that do not appear in, or are poorly represented in response averages, and second, ignores ongoing EEG processes that may be partially time and phase locked by experimental events, thereby contributing portions of response averages (Delorme et al., 2002, Makeig et al., 2002).

In the past decades, pioneer researchers have tried to apply to EEG data analysis techniques developed in electrical engineering and information theory, including time/frequency analysis (Bressler and Freeman, 1980, Makeig, 1993, Neuenschwander and Varela, 1993, Pfurtscheller and Aranibar, 1979, Tallon-Baudry et al., 1996, Weiss and Rappelsberger, 1996) and Independent Component Analysis (ICA) (Jung et al., 2001, Makeig et al., 1996, Makeig et al., 1997, Makeig et al., 1999). These techniques have revealed EEG processes whose dynamic characteristics are also correlated with behavioral changes, though they cannot be seen in the averaged ERP. For example, short-term changes in spectral properties of the ongoing EEG in specific frequency bands may be correlated with cognitive processes, e.g. expectancy of a target stimulus (Makeig et al., 1999) and with visual awareness (Rodriguez et al., 1999). The sufficiency of studying average ERPs has also been questioned by Makeig et al. (2002), who showed that some average ERP peaks may result from partial synchronization of oscillatory EEG processes to time locking events in single data trials.

Currently, most EEG researchers still interpret their data by measuring peaks in event-locked ERP averages. Free availability of more general and easy-to-use signal processing software for EEG data may encourage the wider adoption of more inclusive approaches. Our EEGLAB software toolbox for Matlab (freely available from http://www.sccn.ucsd.edu/eeglab/) allows processing of collections of single EEG data epochs using ICA and spectral analysis as well as data averaging techniques. Using this toolbox, we have demonstrated the advantages of combining ICA, time-frequency analysis, and multi-trial visualization in several publications (e.g., Delorme and Makeig, 2003, Delorme et al., 2002, Makeig et al., 1999, Makeig et al., 2002). In EEGLAB, all these functions are available under a common graphic interface under Matlab, a widely used multi-platform computing environment. EEGLAB extends the collection of publicly available Matlab packages for brain imaging including SPM (Friston, 1995) and FRMLAB (Duann et al., 2002a) for functional MRI studies and Brainstorm (Baillet et al., 1999) for EEG/MEG source analysis.

Section snippets

Basic functions

The ICA/EEG toolbox of Makeig et al. (1997) included a collection of Matlab functions for signal processing and visualization of EEG data including runica(), a function for automated infomax ICA decomposition (Makeig et al., 1997, Makeig et al., 1997), ERP-image plotting (Jung et al., 1999, Makeig et al., 1999), a method of visualizing time-locked potential variations across sets of single trials, and time-frequency decomposition (Makeig, 1993). By 2002, over 5000 researchers from over 50

Discussion

We have developed EEGLAB, a complete interactive environment for processing EEG (or MEG) data under Matlab, to provide both standard and advanced EEG processing functions developed in our own and other laboratories. EEGLAB is strongly oriented towards single-trial visualization techniques, ICA, and event-related time/frequency analysis. Because the software was developed by and for ERP/EEG researchers, we have taken care to make the data processing as transparent as possible and to allow users

Acknowledgements

The authors acknowledge the contributions to EEGLAB and its hundreds of functions by many contributors. Principal among these were Colin Humphries, who wrote the topographic plotting functions and the first version of the data scrolling display function, Sigurd Enghoff, who wrote the first versions of the time/frequency functions and translated the MATLAB-coded runica() infomax ICA function to binary, and Tzyy-Ping Jung, who contributed the first version of the erpimage() function. The runica()

References (41)

  • S Weiss et al.

    EEG coherence within the 13–18 Hz band as a correlate of a distinct lexical organisation of concrete and abstract nouns in humans

    Neurosci. Lett.

    (1996)
  • Anemüller J, Sejnowski TJ, Makeig S. Complex independent component analysis of frequency-domain electroencephalographic...
  • S.-I Amari et al.

    A new learning algorithm for blind source separation

    Adv. Neural Inf. Process. Syst.

    (1996)
  • Baillet S, Mosher JC, Leahy RM, Shattuck DW. BrainStorm: a matlab toolbox for the processing of MEG and EEG signals....
  • A.J Bell et al.

    An information-maximization approach to blind separation and blind deconvolution

    Neural Comput.

    (1995)
  • Bonferroni CE. Sulle medie multiple di potenze. Bollettino dell’Unione Matematica Italiana, 5 third series, 1950. p....
  • J.-F Cardoso et al.

    Blind beamforming for non gaussian signals

    IEE-Proc.-F.

    (1993)
  • A Delorme et al.

    EEG changes accompanying learning regulation of the 12-Hz EEG activity

    IEEE Trans. Rehabilit. Eng.

    (2003)
  • Duann JR, Jung TP, Kuo WJ, Yeh S, Makeig S, Hsieh JC, Sejnowski TJ, Measuring the variability of event-related BOLD...
  • Duann JR, Jung TP, Makeig S, Sejnowski TJ. fMRLAB: an ICA Toolbox for fMRI Data Analysis, In: Human Brain Mapping,...
  • Cited by (16536)

    View all citing articles on Scopus
    View full text