Common spatial patterns using analytic signals for EEG-based BCIs

Abstract

Brain computer interface (BCI) systems provide a control and communication means that depends solely on a user's brain activity. In particular, BCIs based on non-invasive EEG systems can provide a relatively low cost, safe and practical solution compared to other brain imaging techniques, making such systems attractive for use by locked-in patients, for prosthesis control, as well as for gaming applications. A key stage in a typical BCI architecture consists in the association of patterns of mental activity to specific user thoughts and actions. Subsequently, reliable features can be extracted from the recorded EEG signals and through a computer these can be linked to various control and communication functions. A popular technique used to distinguish EEG recordings for a given set of tasks is the method of common spatial patterns (CSP), where spatial filters that optimally discriminate between EEG datasets are determined. In this paper we discuss our development of a variant of the CSP which addresses some limitations of conventional CSP whereby EEG signals are converted into their analytic form. Our CSP-variant – called Analytic CSP (ACSP) – explicitly takes into consideration phase relationships in the multi-channel EEG data thus allowing for disentanglement of amplitude and phase phenomena in the patterns of mental activity. Tests on simulated and real EEG data show that, compared to the standard CSP, the ACSP may improve classification performance and provide a more reliable insight into the underlying brain activity pertaining to a given mental task.