A Dual-path Interactive ECG Denoising Algorithm

Denoising of dynamic electrocardiography (ECG) is a hotspot in ECG processing. Existing denoising methods mainly target single-lead signals or single noise sources, with limited research addressing mixed noise across multiple leads. To address the aforementioned issues, this study proposes a Dual-Path Interactive Denoising Autoencoder (DP-IDAE). The Convolutional Local Denoising Path (CL-Path) of DP-IDAE employs convolutional operations to extract local features of the ECG signal, while the Temporal Global Denoising Path (TG-Path) utilizes bidirectional Long Short-Term Memory networks to capture the signal's temporal global features. Additionally, an Interactive Communication Mechanism (ICM) based on the attention mechanism is proposed to facilitate the exchange of information between the different features of the two paths. The model was tested on denoising performance under different input signal-to-noise ratios for seven types of noise, including single and mixed noise sources comprising baseline wander (BW), muscle artifact (MA), and electrode motion (EM), which are common noise sources in ECG signals. In the MIT-BIH Arrhythmia Database (MLII lead), experimental results demonstrate the superiority of DP-IDAE in ECG signal denoising. Additionally, the study investigates the relationship between denoising effects and signal similarity of different lead ECG, demonstrating excellent cross-lead generalization performance.