Feature-Shuffle and Multi-Head Attention-Based Autoencoder for Eliminating Electrode Motion Noise in ECG Applications

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Electrocardiograms (ECGs) are essential for diagnosing cardiovascular diseases.
• Electrode motion (EM) artifacts significantly degrade ECG signal quality, especially in wearable devices.
• Traditional filtering methods struggle to remove EM artifacts due to frequency overlap with cardiac signals.

Purpose of the Study:

• To develop a novel deep learning model for robust ECG denoising.
• To address the challenge of electrode motion artifacts in ECG signals.
• To improve the accuracy of cardiovascular disease diagnosis in real-world monitoring scenarios.

Main Methods:

• Proposed the Feature-Shuffle Multi-Head Attention Autoencoder (FMHA-AE) architecture.
• Integrated multi-head self-attention (MHSA) to capture long-range dependencies.
• Utilized a feature-shuffle mechanism to enhance representation robustness and generalization.

Main Results:

• FMHA-AE achieved an average signal-to-noise ratio (SNR) improvement of 25.34 dB.
• The model demonstrated a percentage root mean square difference (PRD) of 10.29%.
• FMHA-AE outperformed conventional wavelet-based and deep learning denoising methods.

Conclusions:

• The FMHA-AE model effectively removes electrode motion artifacts while preserving critical ECG morphology.
• This deep learning approach offers a noise-robust solution for ECG analysis.
• FMHA-AE shows significant potential for real-time ECG monitoring in mobile and clinical settings.