Detection of electrocardiogram characteristic points using lifting wavelet transform and Hilbert transform

Abstract

Automatic extraction of time plane features is extremely important for the diagnosis of cardiac diseases. This paper introduces an effective technique based on a second-generation wavelet transform for the denoising of electrocardiogram (ECG) signals. An improved half-soft threshold based on the lifting wavelet is used to overcome the drawbacks of thresholds applied in the classic wavelet. As ECG signals are corrupted by non-stationary noises, after the threshold point of denoising, Hilbert transform and an improved approximate envelope are applied to the ECG signals to enhance the QRS complexes and suppress unwanted P/T waves and noises. A detection algorithm based on the slope threshold is then used for feature detection. The results of simulation demonstrated that the proposed method allows for the detection of the locations of the R peak, P peak, T peak, start and end of the QRS complex, and premature ventricular contraction. The performance of the algorithm was evaluated against the MIT-BIH Arrhythmia Database, and the numerical results indicated that it could function reliably even under poor signal quality and with long P and T peaks. The R peak detection error rate for tape 105 was only 0.27%.