A Common Spatial Pattern Approach for Classification of Mental Counting and Motor Execution EEG

Abstract

A Brain Computer Interface (BCI) as a medium of communication is convenient for people with severe motor disabilities. Although there are a number of different BCIs, the focus of this paper is on Electroencephalography (EEG) as a means of human computer interaction. Motor imagery and mental arithmetic are the most popular techniques used to modulate brain waves that can be used to control devices. We show that it is possible to define different mental states using real fist rotation and imagined reverse counting. While people have already investigated left fist rotation and right fist rotation for dual state BCI, we intend to define a new state using mental reverse counting. We use Common Spatial Pattern (CSP) approach for feature extraction to distinguish between these states. CSP has been prominently used in the context of motor imagery task, we define its applicability for the distinction between motor execution and mental counting. CSP features are evaluated using classifiers like GMM, SVM, and GMM-UBM. GMM-UBM using data filtered through the beta band (13–30 Hz) gives the best performance.