Across-trial averaging of event-related EEG responses and beyond A. Mouraux a , G.D. Iannetti b, ⁎ a Oxford Centre for Functional Magnetic Resonance Imaging of the Brain (FMRIB), Oxford, United Kingdom b Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford, United Kingdom Received 17 December 2007; accepted 14 January 2008 Abstract Internally and externally triggered sensory, motor and cognitive events elicit a number of transient changes in the ongoing electroencephalogram (EEG): event-related brain potentials (ERPs), event-related synchronization and desynchronization (ERS/ERD), and event-related phase resetting (ERPR). To increase the signal-to-noise ratio of event-related brain responses, most studies rely on across-trial averaging in the time domain, a procedure that is, however, blind to a significant fraction of the elicited cortical activity. Here, we outline the key concepts underlying the limitations of time-domain averaging and consider three alternative methodological approaches that have received increasing interest: time-frequency decomposition of the EEG (using the continuous wavelet transform), blind source separation of the EEG (using Independent Component Analysis) and the analysis of event-related brain responses at the level of single trials. In addition, we provide practical guidelines on the implementation of these methods and on the interpretation of the results they produce. © 2008 Elsevier Inc. All rights reserved. Keywords: EEG analysis; Electrophysiology; Event-related potentials (ERPs); Event-related desynchronization (ERD); Event-related synchronization (ERS); Event-related phase resetting (ERPR); Time-frequency analysis; Blind source separation (BSS); Independent component analysis (ICA); Single-trial analysis 1. Electrical brain responses to transient events The ongoing electrical activity of the human brain can be directly sampled through the skull, using one or an array of electrodes placed on the scalp. The recorded electrical acti- vity (the electroencephalogram, EEG) mainly reflects summated, slow post-synaptic potentials of cortical neurons [1]. Sensory, motor or cognitive events (such as a fast-rising sensory stimulus, a brisk self-paced movement or a stimulus- triggered cognitive task) can elicit transient changes in this ongoing electrical activity [2]. However, only a fraction of these changes actually translates into responses that are measurable in the scalp EEG, because the elicited neuronal activity must satisfy a number of conditions to become detectable: (1) The elicited neuronal activity must generate a relatively strong electrical field, and, therefore, it must involve a large population of neurons. (2) The neural activity must be synchronous. Indeed, if the activity is temporally dispersed, the resulting electrical field will be diluted over time, and the signal difficult to measure on the scalp. (3) The activated neuronal population must constitute an open field structure. If its geometrical configuration constitutes a closed field structure (e.g., neurons of a subcortical nucleus, organized in a radially symmetric configuration), the net electrical field outside the active structure will be null, as the electrical fields produced by the neurons of that structure will cancel each other. (4) The time course of the elicited electrical activity must be relatively slow changing, as the skull and scalp interface act as a low-pass filter [3]. Therefore, scalp EEG does not include much of the high- frequency activity evident, for example, in direct intracortical recordings. 1.1. Event-related potentials It is generally accepted that event-related potentials (ERPs) reflect synchronous changes of slow postsynaptic Available online at www.sciencedirect.com Magnetic Resonance Imaging 26 (2008) 1041 – 1054 ⁎ Corresponding author. E-mail address: iannetti@fmrib.ox.ac.uk (G.D. Iannetti). 0730-725X/$ – see front matter © 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2008.01.011