Multichannel high noise level ECG denoising based on adversarial deep learning

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Noise in fetal electrocardiogram (ECG) signals hinders accurate interpretation and limits clinical applications.
• Effective denoising is crucial for reliable non-invasive fetal ECG analysis.
• Existing methods may struggle with complex noise patterns in multi-channel fetal ECG.

Purpose of the Study:

• To develop and evaluate a novel adversarial deep learning-based denoising method for multi-channel fetal ECG signals.
• To enhance the quality and diagnostic utility of non-invasive fetal ECG recordings.
• To improve the accuracy of QRS complex detection in fetal ECG.

Main Methods:

• A three-sub-network adversarial deep learning architecture was employed for end-to-end signal denoising.
• A deep convolutional network with skip connections processed signals from noisy to clean.
• An additional sub-network was utilized to enhance the realism of the denoised fetal ECG signals.

Main Results:

• The proposed method achieved an average signal-to-noise ratio (SNR) improvement of 20 dB across various input SNR levels.
• Significant enhancement in fetal signal quality was observed.
• The method substantially increased true positive QRS complex detections and reduced false positives.

Conclusions:

• Adversarial deep learning offers a powerful approach for denoising multi-channel fetal ECG signals.
• The developed method effectively improves signal quality and diagnostic accuracy in non-invasive fetal ECG.
• This technique holds promise for advancing fetal healthcare monitoring applications.