The Use of Evolutionary Algorithm-Based Methods in EEG Based BCI Systems

The Use of Evolutionary Algorithm-Based Methods in EEG Based BCI Systems

Adham Atyabi (Flinders University, Australia), Martin Luerssen (Flinders University, Australia), Sean P. Fitzgibbon (Flinders University, Australia) and David M W Powers (Flinders University, Australia)
Copyright: © 2013 |Pages: 19
DOI: 10.4018/978-1-4666-2666-9.ch016


Electroencephalogram (EEG) based Brain Computer Interface (BCI) is a system that uses human brainwaves recorded from the scalp as a means for providing a new communication channel by which people with limited physical communication capability can effect control over devices such as moving a mouse and typing characters. Evolutionary approaches have the potential to improve the performance of such system through providing a better sub-set of electrodes or features, reducing the required training time of the classifiers, reducing the noise to signal ratio, and so on. This chapter provides a survey on some of the commonly used EA methods in EEG study.
Chapter Preview


Motor nerves and muscles used in nervous system are traditional communication channel for interaction between brain and computer. A Brain Computer Interface (BCI) is a communication device that bypasses the peripheral nervous system and derives intention directly from brain activity, which it then translates into executable commands. A formal definition for BCI proposed at the first International BCI Meeting (Rensselaerville, New York, 1999) is “A brain-computer interface is a communication system that does not depend on the brain’s normal output pathways of peripheral nerves and muscles” (Wolpaw et al., 2000).

BCI devices typically incorporate stages such as signal acquisition, feature extraction, and classification in their operation. Electroencephalogram (EEG) is one of the commonly used non-invasive techniques in BCI for signal acquisition. EEG records variations of the surface potential from the scalp using some electrodes. The recorded signal is expected to reflect the functional activity of the underlying brain. The EEG signal is a mixture of signals that includes the desired brain activity (the summated signal of millions of cells in the cortex), as well as the heartbeat, eye movement, voluntary and involuntary muscle activity and some possible noise.

Feature extraction stage is used to provide alternative representations of the raw measured signals that help the classifier to better discriminate the set of BCI operations. It is common to use preprocessing stage containing activities such as re-referencing electrodes, demeaning, normalizing, dimension reduction, artifact removal, and so on prior to feature extraction.

Although BCI systems have proven successful for providing new options and communication channels for people with severe neuromuscular disabilities, the use of such systems for controlling or communicating tasks that require high speed and accurate results is still difficult and far from being realized. This is due to complexities such as the complex nature of brain signals, signal smearing and attenuation due to volume conduction, environmental noise and biological contamination, sensitivity of recording equipment, and the difficulty for subjects in achieving and maintaining a requisite brain states. To date, much BCI research has focused strongly on using signal processing to overcome the poor SNR of the EEG signal measured from the scalp.

EEG can be considered as a set of data-points that follow some pattern with a notable degree of randomness. Considering the high level of dynamism in EEG signal, it would be an appropriate suggestion to address the problem of distinguishing these patterns from each other using theories that address dynamic probability optimization. This reasoning guides us toward the Evolutionary Algorithms (EA) in which a performance function called “population generator function” is optimized under the influence of some predefined assessment criteria called “fitness function”. In EA based methods, complexities such as search dynamism, uncertainty, and multi-objective nature of the targeting optimization can be addressed using different types of “population generator” and “fitness” functions. Various evolutionary based search optimization methods exist such as Evolutionary Programming (EP), Evolution Strategy (ES), Differential Evolution (DE), Genetic Algorithm (GA), Genetic Programming (GP), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), etc. Among all, GA, PSO, and ACO methods attracted most attention due to their applicability across different search optimization categories such as self-learning, un-supervised learning, stochastic search, population-based, and behavior-based search. This chapter provides an introduction and some review on these methods and their use in BCI research and discusses their achievements. The chapter is focused on:

Complete Chapter List

Search this Book: