An Impulsive Noise Rejection Filter for Wearable ECG Signal Processing

Abstract

Objective: QRS detection is essential for ECG signal processing. For real-time dynamic ECG, QRS detection is usually performed on a fixed time window, lengths from several to dozens of seconds. However, the unexpected impulsive noise (usually short-term but large amplitude) within the ECG episode is a disaster for QRS detectors. Thus we aimed to propose a new filter to handle this impulsive noise to improve the QRS detection accuracy in wearable ECG measurement. Methods: ECG signals were acquired by the Lenovo Smart-vest, which is a 12-lead wearable ECG collection device, with a sample rate of 500 Hz. The consecutive ECG signals were manually visual-scanned to pick out the episodes including impulsive noises. A fixed time window of 10 s was used for segmenting the ECG episodes. Then, each 10-s ECG episode was processed by Butterworth band-pass filter (0.5–35 Hz). The common Pan & Tompkins (P&T) QRS detector was performed on the filtered signals. A flexible threshold of 60 ms was used to confirm the true positive detection for QRS complex. One hundred episodes with the detection accuracy less than 60% were selected as the test data for the new proposed impulsive noise rejection (INR) filter. The new INR filter was designed with the combination of first order difference, fast Fourier transformation (FFT) and adaptive filtering. Results: Before the INR filtering, the average QRS detection accuracy of the 100 challenging ECG episodes was only 50.62%. As contrast, with the help of the INR filter, P&T detector can achieve a high detection accuracy of 76.32%. Significance: Impulsive noise is a challenging noise existing in the wearable ECG signals. The new designed INR filter can efficiently reject the impulsive noise, and make benefit for the accurate QRS detection in the dynamic environment.