Use this URL to cite or link to this record in EThOS:
Title: Robust PQRST Complex Detection in ECG Signals
Author: Last, Thorsten
ISNI:       0000 0001 3605 3567
Awarding Body: University of Ulster
Current Institution: Ulster University
Date of Award: 2007
Availability of Full Text:
Access from EThOS:
A study of computerised interpretation of the electrocardiogram (ECG) with the main focus on the pre-processing stage of beat detection is presented in this thesis. The accuracy of ECG classification and analysis is strongly dependent on the results of the beat detection process. The purpose of this research is to further improve, both in terms of accuracy and reliability, beat detection in ECG signal processing. A new concept of multi-component based beat detection is presented based on the two different approaches of cross-correlation (CC) and neural networks (NN). The performance of the two multi-component approaches is evaluated and compared with the results of two benchmarking methods: cross-correlation and non-syntactic} beat detection. In addition, a new beat detection algorithm is proposed. This approach is a combination of the two single component and the two multi-component algorithms. Additional improvements of the performance to detect the different waveform components correctly is anticipated by this new approach as it is considered to enhance the advantages of each method and offer a means to suppress their weaknesses. An ECG database containing 100 ECG signals with approximately 3000 cardiac cycles was used to measure the performance of the approaches in correctly detecting the different ECG waveform components. Results have shown the ability of achieving a 7% improvement in detecting individual waveform components when combining the four beat detection methods. Furthermore, the combined approach was able to reduce the number of incorrectly detected waveform components by a factor higher than 15. Extensive evaluation procedures were applied to validate the performance and behaviour of the approaches considering practical influences from possible environments and applications. The beat detection algorithms were evaluated based on applying two different training techniques to the methods, a patient specific and a generic approach, prior to test. Furthermore, the beat detection algorithms were evaluated based on the influence of three different noise sources. Motion artefacts, electromyographic (EMG) interference and possible baseline drifts are three of the most cornmon sources of noise present in EeG recordings. The tests were performed using varying signal to noise ratios (SNR). The presented investigations have consolidated and enhanced the work of previous investigators. These investigations have also resulted in the suggestion of a new technique for the use of beat detection approaches.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available