ECG signal processing for holter monitoring of heart patients is still exploratory. Many signal processing techniques have been evolved for classification and compression of ECG signal. Despite an increase in research in this area, many challenges remain in designing an efficient classification and compression algorithm for ECG signal. These challenges include classification accuracy, good compression ratio with acceptable diagnostic quality etc. This chapter addresses a classification and a compression algorithm based on discrete wavelet transform. Classification algorithm uses discrete wavelet transform based feature to classify abnormal heart beat from ECG signal. Support vector machine is used as a classifier to detect abnormal heartbeat. The compression algorithm utilizes discrete wavelet transform and run-length encoding as a compression tool. Proposed classification and compression algorithms can be employed in monitoring of cardiac patients using holter device.
TopIntroduction
Heart muscles generate tiny electrical impulses which can be recorded by electrodes by placing it on patient’s body surface. This recording is called electrocardiogram (ECG) which is a non-invasive clinical tool in the field of cardiology. It is widely used by cardiologists for diagnosis of heart diseases. ECG waveform of a normal cardiac cycle is depicted in Figure 1 which consist P-wave, QRS-complex and T-wave (Chen et. al, 2014). Depolarization of atrium produces P-wave. Depolarization of right atrium reflects first half of the P-wave while latter half of the P-wave indicates depolarization of left atrium. Q-wave, R-wave and S- wave are collectively known as QRS-complex.
Figure 1. ECG waveform of normal cardiac cycle
It is generated by left and right ventricular depolarization. Ventricular re-polarization generates T-wave. Cardiologists diagnose heart diseases on the basis of different features of ECG waveform which includes amplitudes of PQRST waves and time intervals between different segments. ECG signal is recorded for several hours in holter monitoring of cardiac patients.
Holter Monitor
In pathological ECG test, ECG is recorded for very short duration. For a cardiac patient, it is possible that abnormal heart rhythm may come and go. In general, cardiologist cannot detect abnormal behavior of heart by observing short duration of pathological ECG record. Therefore long time monitoring of cardiac patient is necessary. Holter monitor provides this facility to cardiologist using which observation of heart function is possible on long-term basis. It also helps the cardiologist to determine medicine is working properly or not on cardiac patients. Holter monitor is a battery operated, small device which is used to record ECG signal for 12 hours to 48 hours continuously during normal daily routine of a cardiac patient. Holter monitoring is also called ambulatory ECG monitoring.
Classification of Heartbeats
Classification of heartbeats plays a major role to detect abnormal heartbeat during monitoring of heart patients. Many heartbeat classification techniques have been developed based on its features. These features include morphological features (De Chazal et. al, 2004), segmentation based features (De Chazal et. al, 2004) and frequency domain features (Romero et. al, 2001, De Chazal et. al, 2004). The technique of classification of heartbeats started two decades before using artificial neural network as a classifier (Hu et. al, 1992). At present classifier such as support vector machine (SVM) (Osowski et. al, 2004), hidden Markov model (HMM) (Coast et. al, 1990, Kolekar et. al 2004), artificial neural network (Hu et. al, 1992, Hu et. al, 1997, Barro et. al, 1998), self organizing maps (Lagerholm et. al, 2000) are widely used for classification of ECG heartbeats. A classification technique based on discrete wavelet transform and HMM is proposed in (Andreão et. al, 2006) using which a premature ventricular contraction (PVC) beat can be classified. In this technique Mexican hat mother wavelet is used for feature extraction of ECG heartbeats. A patient specific classification technique based on discrete wavelet transform (DWT) and artificial neural network (ANN) is proposed in (Ince et.al, 2009) which can be used for detection of ventricular ectopic beats (VEBs) and supra-VEBs. For personalized ECG classification, a block based neural network classifier is proposed in (Jiang et. al, 2007). To classify different class of heartbeats, a technique based on SVM and particle swarm optimization (PSO) is proposed in (Melgani et. al, 2008) which can classify 5 classes of heartbeat. Ye et.al, 2012 proposed a classification algorithm using discrete wavelet transform and SVM. He used morphological features as well as dynamic features (R-R interval) of heart-beats to classify 16 class of ECG signal. Despite an increase in research in classification techniques of ECG heartbeats, good classification accuracy in practical scenario is still a challenge for a classifier.