Source localization for brain-computer interfaces

Abstract

Brain-computer interfaces provide a way to operate software without the requirement for physical movement. Electroencephalography (EEG) can be utilized to detect electrical activity in the brain during the execution of certain mental tasks, which can be used as a control signal for an interface. Automatic interpretation of BCI control signals from multichannel EEG data is generally done by application of a classification algorithm from a particular machine learning paradigm. Classification accuracy and overall BCI performance depends on a feature extraction method, which is used to represent the EEG data according to the characteristic features of a chosen BCI control signal. Certain types of control signals used in BCI can be characterized by their spatial properties. Source localization methods can be used to localize electrically active areas of the user’s brain and, hence, represent the EEG signal by its spatial features. This chapter is dedicated to the essential theory related to electromagnetic source localization problem with a particular focus on the family of sparse localization approaches. First we discuss general electromagnetic head modelling methods used to solve the EEG forward problem. Approaches to inverse problem solving, anatomical regularization and application of source localization to BCI are described later in the chapter. Finally we will discuss sparse source localization methods and present relevant simulation results.