Multi-Channel Fetal ECG Denoising With Deep Convolutional Neural Networks

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Non-invasive fetal electrocardiography (ECG) is crucial for continuous fetal health monitoring.
• Fetal ECG signals are often obscured by significant noise, complicating analysis.
• Existing denoising methods face challenges in effectively removing noise while preserving signal integrity.

Purpose of the Study:

• To develop and evaluate a deep learning-based method for denoising multi-channel non-invasive fetal ECG signals.
• To improve the signal-to-noise ratio (SNR) of fetal ECG recordings.
• To assess the effectiveness of the proposed method on both simulated and real-world fetal ECG data.

Main Methods:

• A deep convolutional encoder-decoder network with symmetric skip-layer connections was employed.
• The network learned end-to-end mappings from noisy fetal ECG signals to clean ones.
• Multi-channel signal information was utilized to enhance denoising performance.

Main Results:

• An average SNR improvement of 9.5 dB was achieved for simulated fetal ECG signals across a wide input SNR range.
• The method demonstrated significant quality improvement on real-world noisy fetal ECG signals.
• Multi-channel processing yielded superior performance compared to single-channel approaches.

Conclusions:

• The proposed deep learning method effectively denoises non-invasive fetal ECG signals.
• The technique preserves beat-to-beat morphological variations without requiring prior noise or pulse information.
• This approach offers a promising advancement for reliable fetal health assessment using non-invasive ECG.