R-Peak Detection in ECG Signal Using Yule-Walker and Principal Component Analysis

Proper diagnosis of clinical Electrocardiogram (ECG) is still a challenge. The minor variations in the attributes of ECG signal cannot be examined properly by simple visualization, rather an efficient technique is required to increase the chances of early prediction of the diseases. R-peak detection is one such important attribute. It plays an important role in the detection of Arrhythmias (heart diseases). Proper detection of Arrhythmias using R-peak requires two things: long time recording of the ECG and noise reduction. But, long time recording of ECG requires proper modeling for extracting features from long data records. In this paper, noise reduction is accomplished using a digital bandpass filter (DBPF), since its filtering characteristics are invariant with drift and temperature, and the features are extracted using Yule-Walker (YW) autoregressive modeling technique which is most appropriate for modeling non-stationary signals recorded for long times. So, databases of AHA (American Heart Association), Ventricular Tachyarrhythmia and MIT-BIH Arrhythmia have been investigated. A total of 18 ECG records were made for implementing the proposed methodology using MATLAB R2008b. The feature extraction step is performed by finding AR coefficients on the basis of the selected model order. YW method is applied for finding AR coefficients and Principal Component Analysis (PCA) is used for R-peak detection. In this paper, normal and abnormal signals have been considered during the detection process. The PCA without YW yields a sensitivity of 99.73%, a specificity of 99.80%, a detection rate (DR) of 99.73%, and an accuracy (ACC) of 99.66%, whereas the proposed PCA with YW (PCA + YW) yields a sensitivity (SE) of 99.88%, a specificity (SP) of 99.92%, a detection rate (DR) of 99.90%, and an accuracy (ACC) of 99.81%. Suitable comparisons of the results obtained using the proposed method have been carried out with those obtained using existing methods for illustration.

Automated summary

Proper diagnosis of clinical Electrocardiogram remains a challenge, with R-peak detection playing a crucial role in detecting arrhythmias. This paper focuses on noise reduction and feature extraction to improve accuracy in long-term ECG recordings, using digital bandpass filter for noise reduction and Yule-Walker autoregressive modeling technique for feature extraction. The proposed method, combining PCA with YW, shows improved sensitivity, specificity, and accuracy compared to traditional methods.