Classification of Arrhythmias Using Statistical Features in the Wavelet Transform Domain Annet Deenu Lopez * and Liza Annie Joseph † Dept. Applied Electronics and Instrumentation Rajagiri School of Engineering and Technology Rajagiri Valley, Kakkanad, Kochi, Kerala, India. Email: * annetdeenu@gmail.com, † liza annie@rajagiritech.ac.in Abstract—Computer assisted recognition and classification of ECG into different pathophysiological disease categories is critical for diagnosis of cardiac abnormalities. Evaluation and prediction of life threatening ventricular arrhythmias greatly depend on Premature Ventricular Contraction (PVC) beats. Many studies have revealed that PVCs when associated with myocardial infarction can be linked to mortality. Hence their immediate detection and treatment is crucial for patients with heart diseases. This work focus on improving the automatic diagnosis of PVC arrhythmia from ECG signals. Out of the different methods for ECG analysis, this work adopts sectional analysis of ECG and suitable statistical features in the wavelet transform domain were calculated . These features were utilized to train Support Vector Machines (SVM) classifier and to classify the ECG as normal or with PVC. Advancement of this work is based on an appropriate choice of minimal statistical features which gives better classification in least time. Keywords—ECG, Arrhythmia, Wavelet Coefficients, Statistical Features, Support Vector Machines I. I NTRODUCTION The recognition and classification of the electrocardiogram (ECG) beat is important for diagnosing different cardiac diseases. Electrocardiography is a method that registers electrical activity of heart against time. Instead of manually annotating the ECG signals by experts such as doctors and cardiologists which could take enormous time and efforts, computer assisted automatic detection of the types of ECG signal can be employed. ECG signals are one of the most important sources of diagnostic information and hence their proper acquisition and processing provide an indispensable tool to support medical diagnosis. Acquired signals are affected by noise and call for advanced filtering techniques. It is expected that any computerized interpretation of ECG signals has to be user- friendly, meaning that the results of classification/interpretation could be easily comprehended by a human user. Among the different arrhythmias, PVC also known as Ventricular Premature Beat (VPB) or extra systole, is a form of irregular heartbeat in which the ventricle contracts prematurely[3]. This results in a skipped beat followed by a stronger beat. Individuals experiencing this condition may report a feeling that his or her heart stops after an attack. The depolarization begins in the ventricle instead of the usual place i.e. the sinus node. PVCs can be a useful natural probe since they induce heart rate turbulence whose characteristics can be measured and used to evaluate cardiac functionality [4]. So is the focus of attention in this paper. Several methods for automated arrhythmia detection have been developed in the past few decades to attempt to assist with the monitoring task. These include signal processing techniques such as frequency analysis[8], wavelet transform[2][7], adaptive neuro fuzzy approach[1], support vector machines[9] and artificial neural networks (ANNs)[10]. Reliable detection of life-threatening arrhythmias though extensively studied during the last decades, remains an open problem. In this paper, how PVC heartbeat can be recognized from a normal one by an appropriate feature selection and extraction scheme is shown. The proposed automated method for the classification of cardiac arrhythmias is based on signal preprocessing, feature extraction and classification. The wavelet transform is used to decompose the ECG signal according to scale. Thus allowing separation of the relevant ECG waveform morphology descriptors from the noise, interference, baseline drift and amplitude variation of the original signal. Several researchers have previously used the wavelet transform coefficients at the appropriate scales as morphological feature vectors rather than the original signal time series and achieved good classification performance. Accordingly, in the current paper, the proposed feature extraction technique employs a suitable wavelet transform in order to effectively extract the morphological and temporal information from ECG data and the extracted features are given as input to the classifier. This study aims at the detection of life-threatening PVC arrhythmias using wavelet transform and support vector machines (SVM). II. MATERIALS In this paper, ECG signals are used as basic signals for classification. The annotated ECG records in this study, available at the MIT/BIH (Massachusetts Institute of Technology and Beth Israel Hospital) arrhythmia database[11], have already been used intermittently for evaluating the different classifiers in recent researches. The database contains 48 records, each containing two-channel (modified limb lead II and modified lead VI) ECG signals for 30 min duration selected from 24-hr recordings of 47 individuals. Continuous ECG signals are band pass-filtered at 0.1-100 Hz and then digitized at 360 Hz. The database contains annotation for both timing information and beat class information verified by independent experts. The first 20 records (numbered in the range of 100-124), which include representative samples of routine clinical recordings, are