R-Peak Detection Using Chaos Analysis in Standard and Real Time ECG Databases

Abstract

ObjectivesTimely and accurate R-peak detection is very important for analyzing electrocardiogram (ECG) signal in critical conditions. The main obstacle in observing the correct relation between underlying physiology and features is that there is no specific method to select the features that are needed for diagnosis of a particular heart disease. Therefore, choice of an advanced feature extraction technique is a major concern especially due to non-linear nature of the ECG signal. Material and methodsIn this study, physioNet (standard) and real-time ECG records have been used. During recording, ECG signal is affected by various noises/interferences which create further challenges in ECG signal analysis. Hence, it requires an effective pre-processing, advanced feature extraction and detection techniques. In this paper, independent principal component analysis (IPCA) is used for pre-processing, since it possesses good characteristic of both principal component analysis (PCA) and independent component analysis (ICA). Due to non-linear nature of ECG signals, chaos analysis is applied in feature extraction stage for different ECG databases. The monitoring and wide description of chaotic patterns of heartbeats are prime concerns for cardiologists. Chaos analysis has been used by estimating different attractors against various time delay dimensions. Correct R-peak detection is useful in diagnosing cardiac diseases and performance of the proposed methodology has been evaluated in terms of sensitivity (Se), positive predictivity (PP), and detection error rate (DER) for both PhysioNet (PN DB) and real-time (RT DB) databases. Results-case-I: Without pre-processingIn this case, R-peaks have been detected using chaos analysis+PCA. The proposed method yields Se of 99.91%, PP of 99.93%, and DER of 0.163% for PN DB and Se of 99.77%, PP of 99.83%, and DER of 0.387% for RT DB. Case-II: With pre-processingIn this case, R-peaks have been detected using IPCA+chaos analysis+PCA. The proposed method yields Se of 99.95%, PP of 99.96%, and DER of 0.093% for PN DB and Se of 99.96%, PP of 99.97%, and DER of 0.055% for RT DB. ConclusionThe proposed technique outperforms the other existing works on various selected evaluation parameters even without pre-processing. Hence, the proposed technique has successfully demonstrated its ability to discriminate different types of heartbeats in most of the critical situations. Therefore, there are strong merits in using chaos analysis as a feature extraction method to reduce the incidence of false diagnosis.