Parametric classification of multichannel averaged event-related potentials

This paper focuses on the systematic development of a parametric approach for classifying averaged event-related potentials (ERPs) recorded from multiple channels. It is shown that the parameters of the averaged ERP ensemble can be estimated directly from the parameters of the single-trial ensemble, thus, making it possible to design a class of parametric classifiers without having to collect a prohibitively large number of single-trial ERPs. An approach based on random sampling without replacement is developed to generate a large number of averaged ERP ensembles in order to evaluate the performance of a classifier. A two-class ERP classification problem is considered and the parameter estimation methods are applied to independently design a Gaussian likelihood ratio classifier for each channel. A fusion rule is formulated to classify an ERP using the classification results from all the channels. Experiments using real and simulated ERPs are designed to show that, through the approach developed, parametric classifiers can be designed and evaluated even when the number of averaged ERPs does not exceed the dimension of the ERP vector. Additionally, it is shown that the performance of a majority rule fusion classifier is consistently superior to the rule that selects a single best channel.