Towards adaptive brain-computer interfaces: Improving accuracy of detection of event-related potentials

Abstract

Electroencefalography (EEG) has a wide range of applications in human-computer interaction and in adaptation and personalization of the interfaces. It can be used either as a sensor, e.g., for emotion detection, or as an input device that allows to take actions based on the brain's response to the presented stimuli. For the latter, it is crucial to be able to reliably detect event-related potentials (ERPs), which can be a hard task because of the noise in the signal, especially when using affordable consumer-oriented devices, such as Emotiv Epoc. In the paper, we present a method of EEG signal processing and classification for detection of ERP P300 wave. We particularly focus on the adaptive channel selection and propose to use genetic algorithm combined with linear discriminant analysis to determine the optimal subset of electrodes for signal processing for each individual user. We evaluated our proposed method on a standard data set outperforming the existing approaches even with decreasing size of a training set. In addition, we conducted a user study with Emotiv Epoc device on a standard P300 Speller task in order to compare the results of our method and to find out, whether this device is suitable for P300 detection.