DCPA-SNN, Direct-Coding-Physics-Aware Spiking Neural Network: A Framework for Wearable ECG Denoising Under...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Wearable electrocardiogram (ECG) monitoring offers continuous cardiac assessment but suffers from signal degradation due to noise.
• Existing noise suppression methods for ECG struggle with dynamic noise conditions common in wearable devices.
• Spiking Neural Networks (SNNs) show potential for low-power processing but are underexplored for ECG denoising.

Purpose of the Study:

• To develop an efficient and robust ECG denoising method for wearable devices.
• To address the limitations of current SNNs in handling complex noise artifacts in ECG signals.
• To propose a novel Direct-Coding-Physics-Aware Spiking Neural Network (DCPA-SNN) for wearable ECG denoising.

Main Methods:

• Developed a Direct-Coding-Physics-Aware Spiking Neural Network (DCPA-SNN) integrating direct-coding SNN, channel attention, and residual noise learning.
• Incorporated a physics-aware multi-domain loss function to preserve crucial ECG waveform structures.
• Evaluated the model using MIT-BIH Arrhythmia and Noise Stress Test databases under various single-noise and mixed-noise scenarios (-6 dB to 4 dB SNR).

Main Results:

• DCPA-SNN demonstrated robust denoising performance across diverse noise conditions.
• In mixed-noise scenarios, the average denoised Signal-to-Noise Ratio (SNR) reached 5.80 dB, an improvement of 6.80 dB.
• The R-peak detection rate significantly improved from 90.71% to 95.72% after denoising.

Conclusions:

• The proposed DCPA-SNN offers a promising solution for effective wearable ECG denoising.
• The model's ability to preserve waveform structures and improve R-peak detection is crucial for clinical diagnostics.
• DCPA-SNN shows potential for low-power deployment in wearable health monitoring systems.