An Algorithm for Automatic QRS Delineation Based on ECG-gradient Signal

Abstract

In this work, an algorithm based on digital signal processing and machine learning is developed for QRS complexes detection in ECG signals. The algorithm for locating the complexes uses a gradient signal and the KNN classification method. In the first step, an efficient process for denoising signals using Stationary Wavelet Transform (SWT), Discrete Wavelet Transform (DWT), and a combination of filtering thresholds is developed. In the second stage, the phase of fiducial points detection is carry out, the gradient of the signal is computed for being used as a feature for the detection of the R-peak. Therefore, a KNN classification method is used in order to separate R-peaks and non R-peaks. The algorithm computes a set of thresholds to recalculate the R-peaks positions that has been omitted or falsely detected due to the ECG wave forms. Finally, the each R peak permits locate Q and S peaks. The results indicate that the algorithm correctly detects \(99.7\%\) of the QRS complexes for the MIT-BIH Arrhythmia database and the \(99.8\% \) using the QT Database. The average processing-time that the algorithm takes to process a signal from the denoising stage to fiducial points detection is 4.95 s.