A real-time data mining technique applied for critical ECG rhythm on handheld device

Chin, F 2012, A real-time data mining technique applied for critical ECG rhythm on handheld device, Masters by Research, Electrical and Computer Engineering, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Chin.pdf Thesis Click to show the corresponding preview/stream application/pdf;... 1.70MB
Title A real-time data mining technique applied for critical ECG rhythm on handheld device
Author(s) Chin, F
Year 2012
Abstract Sudden cardiac arrest is often caused by ventricular arrhythmias and these episodes can lead to death for patients with chronic heart disease. Hence, detection of such arrhythmia is crucial in mobile ECG monitoring. In this research, a systematic study is carried out to investigate the possible limitations that are preventing the realisation of a real-time ECG arrhythmia data-mining algorithm suitable for application on mobile devices. Based on the findings, a computationally lightweight algorithm is devised and tested. Ventricular tachycardia (VT) is the most common type of ventricular arrhythmias and is also the deadliest.. A ventricular tachycardia (VT) episode is due to a disorder ofthe regular contractions ofthe heart. It occurs when the human heart ventricles generate a rapid heartbeat which disrupts the regular physiology cycle. The normal sinus rhythm (NSR) of a regular human heart beat signal has its signature PQRST waveform and in regular pattern. Whereas, the characteristics of a ventricular tachycardia (VT) signal waveforms are short R-R intervals, widen QRS duration and the absence of P-waves.

Each type of ECG arrhythmia previously mentioned has a unique waveform signature that can be exploited as features to be used for the realization of an automated ECG analysis application. In order to extract this known ECG waveform feature, a time-domain analysis is proposed for feature extraction. Cross-correlation allows the computation of a co-efficient that quantifies the similarity between two times-series. Hence, by cross-correlating known ECG waveform templates with an unknown ECG signal, the coefficient can indicate the similarities. In previous published work, a preliminary study was carried out. The cross-correlation coefficient wave (CCW) technique was introduced for feature extraction. The outcome ofthis work presents CCW as a promising feature to differentiate between NSR, VT and Vfib signals. Moreover, cross-correlation computation does not require high computational overhead.

Next, an automated detection algorithm requires a classification mechanism to make sense of the feature extracted. A further study is conducted and published, a fuzzy set k-NN classifier was introduced for the classification of CCW feature extracted from ECG signal segments. A training set of size 180 is used. The outcome of the study indicates that the computationally light-weight fuzzy k-NN classifier can reliably classify between NSR and VT signals, the class detection rate is low for classifying Vfib signal using the fuzzy k-NN classifier. Hence, a modified algorithm known as fuzzy hybrid classifier is proposed. By implementing an expert knowledge based fuzzy inference system for classification of ECG signal; the Vfib signal detection rate was improved. The comparison outcome was that the hybrid fuzzy classifier is able to achieve 91.1% correct rate, 100% sensitivity and 100% specificity. The previously mentioned result outperforms the compared classifiers. The proposed detection and classification algorithm is able to achieve high accuracy in analysing ECG signal feature of NSR, VT and Vfib nature. Moreover, the proposed classifier is successfully implemented on a smart mobile device and it is able to perform data-mining of the ECG signal with satisfiable results.
Degree Masters by Research
Institution RMIT University
School, Department or Centre Electrical and Computer Engineering
Keyword(s) ECG
cardiac arrest
ventricular arrhythmias
mobile ECG monitoring
Version Filter Type
Access Statistics: 417 Abstract Views, 1336 File Downloads  -  Detailed Statistics
Created: Fri, 12 Jul 2013, 11:28:59 EST by Brett Fenton
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us