Abstract

Info

Title:: Magnetocardiography signal denoising method based on the GWO-VMD-DLSTM framework

Author(s):: FAN Zhenhua1; LIU Zhiqin2; CHANG Jiaming1; DUAN Junping1; WANG Jiayun1; ZHANG Binzhen1; 1. State Key Laboratory of Extreme Environment Optoelectronic Dynamic Testing Technology and Instruments, North University of China, Taiyuan 030051, China 2. School of Chemistry and Material Science, Shanxi Normal University, Taiyuan 030031, China

Keywords:: Keywords: magnetocardiography signal variational mode decomposition long short-term memory network dual-head attention mechanism signal denoising deep learning

Abstract:: Abstract: To address the challenge that magnetocardiographic (MCG) signals are susceptible to contamination by various noise sources during acquisition, which in turn obscures critical pathological features, a novel denoising algorithm integrating variational mode decomposition (VMD) and a dual-head attention long short-term memory (LSTM) network is proposed. This method utilizes the grey wolf optimization algorithm-optimized VMD to adaptively decompose the raw MCG signal into a series of intrinsic mode functions (IMFs), thereby effectively suppressing mode mixing and achieving precise separation between noise and clinically relevant features. Next, a LSTM network with dual-head attention module is constructed to adaptively capture the noise and signal features on the IMFs, which enables component-wise filtering and end-to-end signal reconstruction without the need for manual IMFs selection. Experimental results demonstrate that in environments with complex noise and baseline drift, the proposed method achieves superior performance compared to conventional algorithms, with a signal-to-noise ratio of 23.58 dB and a cosine similarity of 0.99, and meanwhile effectively preserves diagnostically relevant features. This study provides a robust technical foundation for enhancing the clinical utility of MCG in the early screening and diagnosis of cardiovascular diseases.