R-Peak Detection in Electrocardiogram Signal Using Stationary Wavelet Transform

Abstract

Electrocardiogram (ECG) signal is used to measure the electrical activity of heart as blood circulate from chamber to rest of the body. ECG signal is basically having three waves viz. P-wave, QRS-wave, and T-wave. Among three waves QRS wave is important in which R-wave is responsible for detecting any existing heart disease. Detecting that R-peak with precision help to professional to solve many real-life problems. The accurate R-peak is utmost important for detecting heart diseases before any casualty. The proposed work is utilizing Stationary Wavelet Transform (SWT) for signal pre-processing followed by a Random Forest (RF) classifier to identify R-peak candidates. In SWT, a Symlet 4 wavelet is used in which decomposition levels can jointly perform signal denoising and extraction of the QRS band energy. It helps to precisely determine R peak candidates by the use of adaptive energy thresholding. The proposed work efficiently applied physiology inspired post processing to accurately confirm identified R peaks. After getting the R-peaks, model is trained and 10 features are evaluated that shows the property of the waveform. From candidate peaks, window size of 20ms is selected and then extracted all 10 features for that. These features are then put into RF to increase the efficiency. Further, fine tuning of the parameters is done and repeat this step until whole signal is trained. The proposed algorithm obtained higher values of performance evaluating parameters viz. an overall classification accuracy (ACC) of 97.54%, sensitivity (SEN) of 98.41%, positive predictive value (PPV) of 98.54%, and F1-score (F1-S) of 98.47% those are higher than other benchmark algorithms.