Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Adaptive techniques for the detection and localization of event related potentials from EEGs using reference signals

Spyrou, Loukianos 2008. Adaptive techniques for the detection and localization of event related potentials from EEGs using reference signals. PhD Thesis, Cardiff University.

[img] PDF - Accepted Post-Print Version
Download (5MB)

Abstract

In this thesis we show the methods we developed for the detection and localisation of P300 signals from the electroencephalogram. We utilised signal processing theory in order to enhance the current methodology. The work done can be applied both to EEG averages and single trial EEG data. We developed a variety of methods dealing with the extraction of the P300 and its subcomponents using independent component analysis and least squares. Moreover, we developed novel localisation methods that localise the desired P300 subcomponent from EEG data. Throughout the thesis the main idea was the use of reference signals, which describe the prior information we have about the sources of interest. The main objective of this thesis is to utilize adaptive techniques, namely blind source separation (BSS), least squares (LS) and spatial filtering, in order to extract the P300 subcomponents from the electroencephalogram (EEG) with greater accuracy than the traditional methods. The first topic of research, is the development of constrained BSS and blind signal extraction (BSE) algorithms, to enhance the estimation of the conventional BSS and BSE algorithms. In these methods we use reference signals as prior information, obtained from real EEG data, to aid BSS and BSE in the extraction of the P300 subcomponents. Although, this method exhibits very good behaviour in terms of EEG averaged data, its performance degrades when applied to single trial data, which is the response of the brain after one single stimulus. The second topic deals with single trial EEG data and is based on least squares. Again, we use reference signals to describe the prior knowledge of the P300 subcomponents. In contrast to the first method, the reference signals are Gaussian spike templates with variable latency and width. The target of this algorithm is to measure the properties of the extracted P300 subcomponents and obtain features that can be used in the classification of schizophrenic patients and healthy subjects. Finally, the idea of spatial filtering combined with the use of a reference signal for localisation is introduced for the first time. The designed algorithm localises our desired source from within a mixture of sources where the propagation model of the sources is available. It performs well in the presence of noise and correlated sources. The research presented in this thesis paves the path in introducing adaptive techniques based on reference signals into ERP estimation. The results have been very promising and provide a big step in establishing a foundation for future research.

Item Type: Thesis (PhD)
Status: Unpublished
Schools: Engineering
Subjects: T Technology > TA Engineering (General). Civil engineering (General)
ISBN: 9781303212949
Date of First Compliant Deposit: 30 March 2016
Last Modified: 12 Feb 2016 23:12
URI: http://orca-mwe.cf.ac.uk/id/eprint/54694

Actions (repository staff only)

Edit Item Edit Item