City Research Online

Abstract

Despite recent advances in brain monitoring techniques, the electroencephalogram (EEG) is still widely used in the diagnosis and monitoring of epilepsy. To increase its effectiveness, long-term monitoring of patients was proposed but the large volume of recorded EEG signals produced, made their traditional interpretation by human experts difficult and automatic EEG analysis was proposed as an alternative.

This Thesis is concerned, primarily, with the on-line detection of epileptic transients (spikes) in the interictal EEG signals of patients. A review of previous methods, revealed that the limited success of automatic analysis systems was linked to the vagueness of neurophysiological definitions and the subjectiveness of human interpretation, which is based on experience.

To address these issues, it was realized that a common point of reference is required for the integration of medical and signal processing expertise, which could be provided by a model of the signal. Early attempts to develop such a model are described. These led to the development of spike detectors based on the derivatives of the EEG.

Later, by describing medical definitions with signal processing terminology, a comprehensive model of the signal was constructed. This was based on its decomposition into background activity, spikes, transients and noise and describing each one of them in terms of simple, random signals and quasi-linear systems.

This suggested a method of analysis based on inverse modelling for the decomposition of the EEG. The model for transients was estimated off-line. An on-line system, consisting of adaptive prediction error systems, constrained all-pole adaptive systems and a basic signal detection procedure was implemented. Several alternative adaptive realizations were investigated.

The spike detection procedure was generalized for the detection of other transients. Finally this procedure was replaced by a Multi-Layer Perceptron neural network, whose inherent ability to learn by example is important, as it provides the means to incorporate medical experience without requiring its explicit quantification. The system is flexible and its extension to detect any number of transients is demonstrated. The method may be applied to other signals and improved by new developments in signal processing.

Publication Type:	Thesis (Doctoral)
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science R Medicine > RZ Other systems of medicine
Departments:	School of Science & Technology > Computer Science School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads