AbstractThe electrocardiogram is a skin surface measurement of the electrical activity of the heart over time. This activity is detected by electrodes attached to the surface of the skin and recorded or displayed by an external medical device. Doctors use electrocardiograms to detect and diagnose conditions such as arrhythmias (abnormal heart rhythms) and myocardial infarctions (heart attacks).
The work described in this thesis investigates the system designed for two primary applications, electrocardiogram classification system based on autoregressive models which identifies normal (healthy) from abnormal (unhealthy) electrocardiogram signals and the electrocardiogram biometric system based on analytic and modeling features which identifies each person individually from his or her electrocardiogram. In recent years, a number of signal processing techniques have been used to design electrocardiogram signal auto-classification and biometric identification systems. electrocardiogram classification and biometric systems implemented in this thesis are compared with a number of other recently described techniques and methods to identify electrocardiogram signals. The aim of any designed electrocardiogram classification and biometric system described in this work is to achieve high accuracy rate when identifying electrocardiograms. Electrocardiogram classification and biometric systems consists of four major stages, pre-processing of electrocardiogram signal, QRS complex detection, feature extraction and classification algorithms.
Each of those steps are discussed and explained in separate chapters with variety of techniques and methods employed to achieve each step. Developed systems based on autoregressive models to design electrocardiogram classification and biometric achieved accurate correct classification rate with high level of productivity due to the small number of extracted parameters using autoregressive models. The proposed electrocardiogram classification and biometric systems of this work achieved 100 % correct classification rate in identifying normal from abnormal electrocardiogram signals and each person individually from his or her electrocardiogram signal. In this work, it has been proven that autoregressive models can represent electrocardiogram signals with 91% accuracy and matching between the original electrocardiogram signal and the modeled signal.
|Date of Award||Jun 2015|
|Supervisor||Branislav Vuksanovic (Supervisor)|