QRS complex recognition based on adaptive wavelet threshold and Hilbert transform for continuous ECGs

Abstract

The wearable continuous ECG monitoring and cardiovascular disease detection system has the characteristics of strong noises and big continuous data. It puts forward higher requirements on the accuracy and efficiency of the QRS recognition algorithm. The currently used QRS detection algorithms still have the problem of missed detection and false detection for continuous ECG data. Therefore, a fast R-peak recognition method based on self-adaptive wavelet threshold and Hilbert transform is proposed for processing noisy continuous ECGs. First, wavelet self-adaptive threshold filter is used to denoise the dynamic ECGs, and then the first-order difference, Shannon energy envelope extraction, Hilbert transform and self-adaptive threshold back-check technology are employed for R-peak detection. Using the proposed method, experimental results show that the accuracy, sensitivity, and specificity of the MIT-BIH arrhythmia database are 99.79%, 99.92%, and 99.87%, respectively. Using the PhysioNet/CinC Challenge 2014 dynamic ECG database, the accuracy, sensitivity and specificity are 96.89%, 97.92% and 98.92%, respectively. Furthermore, the realtime ECGs collecting from the portable ECG detector mECG-101 are also used to evaluate the method. The experimental results show that the accuracy, sensitivity and specificity reach to 97.75%, 98.25% and 99.47%, respectively. Compared with Pan-Tompkins algorithm and wavelet transform algorithm, the proposed method has higher detection accuracy and generalization ability, especially for the wide and low-amplitude QRS complexes.