DeScoD-ECG: Deep Score-Based Diffusion Model for ECG Baseline Wander and Noise Removal.

Area of Science:

• Biomedical Signal Processing
• Artificial Intelligence in Healthcare
• Cardiovascular Diagnostics

Background:

• Electrocardiogram (ECG) signals are crucial for diagnosing heart conditions.
• Noise interference, particularly baseline wander, degrades ECG signal quality and diagnostic accuracy.
• Effective noise removal is essential for reliable ECG interpretation.

Purpose of the Study:

• To develop and evaluate a novel technology for removing baseline wander and noise from ECG signals.
• To enhance the fidelity and quality of reconstructed ECG signals for improved diagnostic capabilities.
• To introduce a new deep learning-based approach for ECG signal denoising.

Main Methods:

• Extended a diffusion model into a conditional approach named Deep Score-Based Diffusion model for Electrocardiogram baseline wander and noise removal (DeScoD-ECG).
• Implemented a multi-shots averaging strategy to further enhance signal reconstruction quality.
• Validated the method on the QT Database and MIT-BIH Noise Stress Test Database, comparing against traditional and deep learning baseline methods.

Main Results:

• The proposed DeScoD-ECG method demonstrated superior performance across four distance-based similarity metrics.
• Achieved at least a 20% overall improvement compared to the best-performing baseline methods.
• Showcased enhanced stability and approximation of true data distribution, even under significant noise.

Conclusions:

• DeScoD-ECG represents a state-of-the-art solution for ECG baseline wander and noise removal.
• The model exhibits improved data distribution approximation and stability under challenging noise conditions.
• This innovative application of conditional diffusion models holds significant potential for widespread use in biomedical applications.