GMM-based Detection Methods in EEG-based Brain-Computer Interfaces

Abstract

Brain Computer Interface (BCI) technology allows a person to control a device by bypassing the use of muscular activity. In previous studies, Signal processing and classification methods play a decisive role in the performance accuracy in BCI application, and there is a current requirement for preliminary analyses to identify the brain signal features best suited for communication. This paper designs proposal of signal detection instead of classification, and which does not require such perplexing analysis procedures. This signal detection concept is carried out by establishing Gaussian mixture models (GMM) of resting brain activity (idle state), so that any imagined movement or real movement signals in brain can be detected. GMM as a versatile modeling tool can be used to approximate any probability density function (pdf) given a sufficient number of components, and impose only minimal assumptions about the modeled random variables. Meanwhile GMM is effective in the calculation. Our best results were 82% on BCI Competition 2002 - Data set and 75% on BCI Competition 2003 - Data set IV in accuracy rate. It confirms the truth of feasibility on detection. In summary, this paper demonstrates how the approach of detection could be used to overcome one of the present impediments to translation of laboratory BCI demonstrations into clinically practical applications.