Electrocardiogram Signal Denoising Based on Multi-Threshold Stationary Wavelet Transform

With the increasing risks of cardiovascular diseases (CVDs) all over the world, electrocardiogram (ECG) monitoring has become an important means for the timely diagnosis of CVDs. However, ECG signal can be easily disturbed by noises such as motion artifact (MA) when recorded by wearable devices in our daily life. To eliminate these noises in ECG signal, a denoising algorithm based on multi-threshold stationary wavelet transform (SWT), called MT-SWT, is proposed. We first propose a QRS complex detection algorithm based on joint threshold judgement to accurately separate the QRS complex from the other waves of ECG signals. Then, taking historical ECG signals when the human body is static as the reference signals, we set multiple thresholds for different SWT coefficients and different parts of ECG signals respectively. Finally, for a section of the input ECG signal, each SWT coefficient is processed by a given soft thresholding function for denoising. We compare MT-SWT with other algorithms based on MIT-BIH datasets, and also implement it in real-world ECG monitoring wearable devices. The experimental results show that compared with the state-of-the-arts, MT-SWT achieves higher accuracy on QRS complex detection under the condition of low signal-to-noise ratio (SNR). Moreover, MT-SWT achieves high SNR improvement ($SNR_{imp}$) and low percent root mean square difference ($PRD$) under different SNR conditions.