QRS Detection Using Zero Crossing Counts

The importance of QRS detection results from the wide use of the timing information of this component, e.g., in heart rate variability analysis, ECG classification, and ECG compression. In most cases, the temporal location of the R-wave is taken as the location of the QRS complex. Missed or falsely detected beats are problematic in all of these applications and may lead to poor results. The literature [1,2] clearly indicates that the number of false detections may increase significantly in the presence of poor signal-to-noise ratios or pathological signals. Detection errors can be reduced by the application of computationally more expensive algorithms, for instance by the implementation of reverse search methods. However, particularly in the case of battery-driven devices, the computational complexity needs to be kept low. Hence, a tradeoff between computational complexity and detection performance needs to be found. The detection of QRS complexes and R-waves in ECG signals has been studied for several decades. Most of the earliest algorithms are based on feature signals obtained from the derivatives of the ECG signal [3-9]. As long as no additional rules for the reduction of false detections are applied, these methods are characterized by low computational complexity and relatively poor detection results in the presence of problematic signals (e.g., containing baseline drift, noise and artifacts, as well as changes in the QRS morphology). For an overview of algorithms based on the first and second order derivative of the ECG, see [1]. Additionally, more sophisticated digital filters have been used as peak detectors for ECG signals, providing better feature signals for disturbed ECG signals [2,919]. Many standard signal processing methods have been applied to QRS and R-wave detection, such as methods from the field of linear and nonlinear filtering [20], wavelet transform [21-22], artificial neural networks [23-24], genetic algorithms [25], and linear prediction [26]. These algorithms [1,20-28] are generally much more complex compared to derivative-based methods and thus exhibit significantly better detection results. In this paper, an algorithm is proposed that simultaneously meets the demands of a low computational load QRS Detection Using Zero Crossing Counts