Fully-Gated Denoising Auto-Encoder for Artifact Reduction in ECG Signals.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Cardiovascular diseases (CVDs) are a leading cause of global mortality.
• Accurate ECG signal analysis is vital for CVD diagnosis, but ambulatory ECGs are prone to artifacts.
• Existing ECG denoising methods often fail to preserve signal morphology, especially under high noise conditions.

Purpose of the Study:

• To develop a novel Fully-Gated Denoising Autoencoder (FGDAE) for robust ECG denoising.
• To significantly reduce the impact of various artifacts on ECG signal quality.
• To achieve maximal morphological preservation of ECG signals during the denoising process.

Main Methods:

• Proposed a FGDAE model incorporating gating mechanisms in all layers and skip connections.
• Utilized Self-organized Operational Neural Network (self-ONN) neurons within the encoder.
• Developed a multi-component loss function for efficient latent representation learning and denoising.

Main Results:

• FGDAE demonstrated superior performance across seven error metrics compared to state-of-the-art algorithms.
• The model achieved reliable denoising even in extreme noise conditions and complex artifact mixtures.
• FGDAE offers significant model size reduction (61-73%) and improved inference speed.

Conclusions:

• The proposed FGDAE provides highly effective ECG denoising with excellent morphological preservation.
• FGDAE shows practical benefits for real-world applications due to its efficiency and reduced size.
• Further research is needed for optimal preservation against specific artifacts like electrode motion.