An Application of Morphological Feature Extraction and Support Vector Machines in Computerized ECG interpretation Wai Kei Lei, Bing Nan Li, Ming Chui Dong, Bin Bin Fu Department of Electrical & Electronics Engineering, University of Macau, Taipa, Macau Institute of Systems and Computer Engineering of Macau, Taipa 1356, Macau vincelei@hotmail.com ; bingoon@ieee.org Abstract This paper presents a novel approach that recognizing heart rhythm with the combination of adaptive Hermite decomposition and support vector machines (SVM) classification. The novelty lies in two aspects. In the first aspect, for the goal of feature extraction, the orthogonal transformation based on Hermite basis functions is proposed to characterize the morphological features of ECG data. In the other aspect, as to the multi-class electrocardiogram (ECG) classification, the one-against-all strategy is applied to a cluster of binary SVMs. Finally, in terms of numerical experiments, the major types of heart rhythms in the MIT-BIH arrhythmia database are taken into account. The results confirm its reliability and accuracy of the proposed ECG interpreter. 1. Introduction In modern society, cardiovascular disease has been one of the leading causes of death for years [1]. At the same time, early prevention and risk reduction are confirmed as the most effective way of treatment. The electrocardiograms (ECG) from cardiac health monitoring reflect the subtle procedure of myocardial depolarization and repolarization in depth. So nowadays ECG monitoring and analysis have been recommended as one of golden methods in clinical medicine for cardiovascular disease screening. However, for any cardiologist, it is an oppressive while time-consuming job to inspect and analyze ECG recordings. As a consequence, the computers and information technology are naturally introduced for ECG interpretation [2]-[7]. In essence, ECGs are the graphical representation of myocardial bioelectrical activities. A typical ECG waveform consists of a QRS complex (the composite of Q, R and S waves), a P and a T wave. Among them, the QRS complex is generally remarkable and most significant. So, as a common, the cardiologists pay more attention on the morphological features of ECG QRS complex in order to determine a normal or abnormal ECG beat. Over the past years, a good many of methodologies and algorithms have been proposed to interpret ECG recordings in stead of those cardiologists. Both clinical as well as numerical experiments have confirmed the competency of computerized ECG interpretation [4]. In general, the computerized ECG interpretation involves two independent modules: the first one is for feature extraction and pattern definition; and the second one is for classification or clustering. Different feature sets have been evaluated systematically for ECG Sixth Mexican International Conference on Artificial Intelligence, Special Session 978-0-7695-3124-3/08 $25.00 © 2008 IEEE DOI 10.1109/MICAI.2007.32 82