An Efficient Electrode Ranking Method for Single Trial Detection of EEG Error-Related Potentials

Abstract

The human brain's response to mistakes or erroneous events is termed as Error-Related Potential (ErrP). The ErrP can be recorded non-invasively using Electroencephalogram (EEG). The ErrP activity is localized and gets reflected in a few EEG electrodes only. Further, EEG offers a poor signal-to-noise ratio. Therefore, single-trial detection of ErrP from EEG data is challenging. The objective of this work is to propose an efficient method for selecting electrodes that carry ErrP related information to enhance single-trial detection accuracy. In this work, the cosine similarity and Euclidian distance measures are used to rank the EEG electrodes. The selected top-ranked electrodes are used to extract electrode-average features followed by a classifier. This work is implemented on a public dataset containing 6 subjects' datasets each having 2 sessions of EEG data. The two proposed electrode ranking methods - cosine similarity measure and Euclidian distance measure are implemented separately. Both electrode ranking methods aided in achieving equally good ErrP detection rates. The cross-validated average detection rates achieved using the proposed electrode ranking methods are ~73.5% and ~80% for error and correct trials respectively. Further, the results are compared with three existing methods including Convolutional Neural Network (CNN) implemented on the same dataset used in this work to show the efficiency of the proposed method. The significance of this work is that the single-trial detection of ErrP can aid in improving the classification accuracy of decoding EEG tasks in Brain-Computer Interface systems.