Time Resolved Multivariate Pattern Analysis for Infant EEG Data
Time-resolved multivariate pattern analysis (MVPA), a technique for analyzing magneto- and electro-encephalography (M/EEG) data, quantifies the extent and timecourse of neural representations that support the discrimination of stimuli. Because MVPA is yet to be widely applied in infant research, there is a need to develop best practices that account for the limitations of infant neuroimaging (i.e., limited data, noise). We assessed MVPA analysis approaches with infant EEG data, including manipulating trials per stimulus, and preprocessing procedures. MVPA was performed to examine the timecourse of infants’ neural representations supporting image discrimination, resulting in significantly distinct neural representations for the images against a chance level of 50%, as well as an empirical chance generated by repeated permutation of the data. Geometric and accuracy-based Representational Dissimilarity Matrices (RDMs) were additionally generated to characterize the space of neural representations. In this work we investigate best practices for MVPA with infant EEG such as preprocessing, determining valid trial thresholds, and classification algorithms, and outline issues when there are deviations from these recommendations.