Thesis

Event detection in EEG signals for brain computer interface using expectation-maximation algorithm

Downloadable Content

Download PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2013
Thesis identifier
  • T13661
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Past two decades have witnessed a steady growth in research related to Brain Computer Interface (BCI), which offers a non-muscular communication pathway to patients disabled due to neurological disorders. BCI works by recording brain signals (example; electroencephalography (EEG)) and translating them into machine-understandable language. Most of the current BCI systems identify features of the brain signals and classify them according to a predefined criterion set by the classifying algorithm. The features of the brain signals can change over time and this could adversely affect the feature extraction and classification algorithm. This restricts the use of BCI in a laboratory oriented device. Due to these impediments, this study aims at detecting events in EEG signals rather than classifying them. As detection would not require classification of the features, the process is less susceptible to changes in signal features. Thus, it could be possible to bring BCIs into clinical use. The current study modelled the features of rest EEG signal using a Gaussian distribution and a mixture of Gaussians. The features of EEG were extracted using continuous wavelet transform and the parameters of the models were estimated using an Expectation-Maximization (EM) algorithm. Maximum-likelihood estimates of the parameters of the rest EEG and EEG during motor actions were compared. Statistical analysis of the results indicates that there are differences between maximum-likelihood values of the wavelet coefficients of the rest EEG and EEG undergoing motor activity. Two models of Gaussian mixture gave better results than a simple Gaussian distribution. This shows that Gaussian mixture modelling (GMM) of EEG is sensitive enough to depict changes during motor activity with respect to rest EEG.
Resource Type
DOI
Date Created
  • 2013
Former identifier
  • 1005037

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PDF of thesis T13661 2021-07-02 Public Download