Article Preview
Top1. Introduction
An electrocardiogram (ECG) is a record of the variations in the bioelectric potential with respect to time as the human heart beats (Figure 1). An ECG is an important tool during the diagnosis of heart conditions. It provides valuable information about the functional aspects of the heart and the cardiovascular system. The topic of QRS detection has been widely investigated with regard to its practical importance in the field of clinical medicine. For example, QRS detection yields an important basis for the instantaneous heart rate (HR) computation since the accuracy of the instantaneous heart rate estimation relies on the performance of QRS detection (Köhler et al., 2002; Chan et al., 2005; Gűler & Űbeyli, 2005). The QRS complex of an ECG represents ventricular depolarization, causing the ventricles to contract; this can be used to determine the patient’s cardiac rate. An ECG wave has a time-varying morphology and can be influenced by physiological variations in different patients and corruption because of noise. Since the large number of examinations performed yields variegated information, visual evaluation of the data becomes difficult. Moreover, manual analyses are prone to errors because of visual fatigue.
Figure 1. Schematic diagram and feature points of one cardiac cycle in a typical ECG signal
Because of its clinical importance, numerous approaches for QRS detection have been already proposed (Chen et al., 2006; Paoletti & Carlo, 2006; Martinez et al., 2004; Hamilton & Tompkins, 1986; Pan & Tompkins, 1985). Some existing QRS detection algorithms employ a specific QRS template; this might be considered the best way to prevent degradation in the QRS detection performance by undesired noise sources, including baseline drifts, artifacts induced by electrode motion and power-line interference (Köhler et al., 2002; Guglin & Thatai, 2006). There exist some familiar R-point detection algorithms, which can be divided into three categories: (1) algorithms based on amplitude and first derivative, (2) algorithms based on first and second derivative and (3) algorithms based on digital filters. However, some of them are extremely complex for real-time processing of data (Gray et al., 1990).
Many approaches for QRS complex detection involve a pre-processing stage where the ECG signal is transformed to accentuate the QRS complex and a decision stage where a QRS complex is detected using thresholding. The most popular method is the Pan-Tompkins (PT) algorithm (Pan & Tompkins, 1985). The PT algorithm can be combined with the wavelet transformation (WT) method to benefit from the advantages of both these methods and parameters can be introduced such that the contribution of the individual algorithms can be balanced; these parameters can be estimated in a data-driven manner (Meyer et al., 2006). In another method, QRS segmentation is based on the combination of WT and adaptive threshold, and the QRS complex can be identified without the preprocessing stage (Madeiro et al., 2007). Signal segmentation is based on the QRS detector, which is specifically designed for noisy applications (ambulatory recordings), and the Karhunen-Loève (KL) transform provides an efficient reduction in the parameter space dimension (Paoletti & Carlo, 2006). Adaptive wavelets integrated with neural network are used in the compression and classification of ECG signals. The wavelet parameters and the relative importance (weight) of each basis function can be approximately optimized with respect to a given function (signal) in the minimum mean square error (Kadambe & Srinivasan, 2006).