Abstract
In recent years, time-resolved multivariate pattern analysis (MVPA) has gained much popularity in the analysis of electroencephalography (EEG) and magnetoencephalography (MEG) data. However, MVPA may appear daunting to those who have been applying traditional analyses using event-related potentials (ERPs) or event-related fields (ERFs). To ease this transition, we recently developed the Amsterdam Decoding and Modeling (ADAM) toolbox in MATLAB. ADAM is an entry-level toolbox that allows a direct comparison of ERP/ERF results to MVPA results using any dataset in standard EEGLAB or Fieldtrip format. The toolbox performs and visualizes multiple-comparison corrected group decoding and forward encoding results in a variety of ways, such as classifier performance across time, temporal generalization (time-by-time) matrices of classifier performance, channel tuning functions (CTFs) and topographical maps of (forward-transformed) classifier weights. All analyses can be performed directly on raw data or can be preceded by a time-frequency decomposition of the data in which case the analyses are performed separately on different frequency bands. The figures ADAM produces are publication-ready. In the current manuscript, we provide a cookbook in which we apply a decoding analysis to a publicly available MEG/EEG dataset involving the perception of famous, non-famous and scrambled faces. The manuscript covers the steps involved in single subject analysis and shows how to perform and visualize a subsequent group-level statistical analysis. The processing pipeline covers computation and visualization of group ERPs, ERP difference waves, as well as MVPA decoding results. It ends with a comparison of the differences and similarities between EEG and MEG decoding results. The manuscript has a level of description that allows application of these analyses to any dataset in EEGLAB or Fieldtrip format.
Highlights
Since Haxby and colleagues popularized multivariate pattern analysis (MVPA) for functional magnetic resonance imaging (Haxby et al, 2001), multivariate approaches have gained widespread popularity
Multivariate analysis in EEG and MEG offers a number of analytical advantages over univariate time-series analysis, such as the ability to look at temporal generalization to characterize neural dynamics over time (King and Dehaene, 2014), the use of representational similarity analysis to map different physiological measures or anatomical substrates onto each other (Kriegeskorte et al, 2008; Cichy et al, 2014), as well as the ability to establish a common performance measure to map behavioral onto neural data (Fahrenfort et al, 2017b)
The results show that event-related potentials (ERPs) can show similar outcomes as decoding analyses, as long as one knows which electrode(s) to select
Summary
Since Haxby and colleagues popularized MVPA for functional magnetic resonance imaging (fMRI) (Haxby et al, 2001), multivariate approaches have gained widespread popularity. The multivariate nature of EEG has long been recognized (e.g., Peters et al, 1998; Mitra and Pesaran, 1999), widespread adoption of MVPA to decode experimental conditions using brain activity has been much slower in EEG and MEG research than in fMRI. Multivariate analysis in EEG and MEG offers a number of analytical advantages over univariate time-series analysis, such as the ability to look at temporal generalization to characterize neural dynamics over time (King and Dehaene, 2014), the use of representational similarity analysis to map different physiological measures or anatomical substrates onto each other (Kriegeskorte et al, 2008; Cichy et al, 2014), as well as the ability to establish a common performance measure to map behavioral onto neural data (Fahrenfort et al, 2017b). Many researchers prefer to use multivariate analyses over traditional ERP/ERF analyses based on signals averaged over epochs (Mostert et al, 2015; Kaiser et al, 2016; Wardle et al, 2016; Contini et al, 2017; Marti and Dehaene, 2017; Turner et al, 2017)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
