Application of wavelet-based maximum entropy on the mean in channel optimization for BCI

Abstract

Localizing event-related cortical sources is a key factor while developing a computationally efficient Brain Computer Interface (BCI). This paper proposes a unified application of wavelet-based Maximum Entropy on the Mean (wMEM), as a channel selection method, for classifying two motor imagery (MI) tasks using optimal electroencephalography (EEG) sources. The EEG data, which are collected from publicly available BCI Competition III, are captured from five healthy individuals. This source optimization tool has been validated with a generic BCI framework, which utilizes common spatial pattern with and without regularization as preprocessing tools. However, the best classification accuracy attained is 98% using only 11 selected channels that is close to 100% attained using available 118 channels. This result summarizes how optimal EEG channels can be used to develop a BCI system without compromising the performance significantly.