Student:	Ramazan KOÇOĞLU
Supervisor:	Assoc. Prof. Dr. Temel KAYIKÇIOĞLU
Department:	Elektrik-Elektronik Müh.
Institution:	Graduate School of Natural and Applied Sciences
University:	Karadeniz Technical University, Turkey

Title of the Thesis:	Efficient Classification Of Imagery Electrocorticography (ECoG) Signals In Brain Computer Interface Applications
Level:	M.Sc.
Acceptance Date:	10/8/2009
Number of Pages:	89
Registration Number:	i2083

Summary:

A brain computer interface (BCI) is a communication system that does not depend on

the normal output pathways consisting of periphery nerves and muscles. BCI transforms

mental intentions into control commands by analyzing the biomedical brain activity. The

technique can be an interface to patients who totally losing volitional motor ability like

amyotrophic lateral sclerosis (ALS). Thus, life quality of these type patients will be

increased.

In the past decade BCI technology has developed rapidly. In BCI applications

classification accuracy and information transfer rate are two important issues. The goal in

area of BCI research is to develop a method which has higher classification rate and brain

computer interfacing data rate than existing methods. One method to boost classification

accuracy is to improve the quality of input signal of a BCI system. Electrocorticographic

(ECoG) recordings, derived from surface of the cortex, have the advantages of higher

signal-to-noise ratio and better spatial resolution, and thus may be used as a feasible

alternative of BCI signal source.

In this thesis, it was studied on ECoG dataset which was obtained under different

mental and visual tasks used in literature. In this data set, creating algorithms for

contributing the classification of signals in different session situations is required. Feature

extraction process is the main and also most difficult issue in BCI applications. In this

work features are extracted from ECoG signals which have two different classes by means

of wavelet transform. According to discovered features classification was done by using knearest

neighbor (KNN), support vector machines (SVM) and linear discriminant analyses

(LDA) classifier. All these studies actualized by considering the goal of to obtain higher

classification rate and high brain computer interfacing data rate.

Key Words: ECoG, BCI, Feature Extraction, Classification