[Paper Review] DeScoD-ECG: Deep Score-Based Diffusion Model for ECG Baseline Wander and Noise Removal

Huayu Li, Gregory Ditzler+2 more|arXiv (Cornell University)|Jul 31, 2022

ECG Monitoring and Analysis6 citations

TL;DR

This paper proposes DeScoD-ECG, a conditional score-based diffusion model that iteratively denoises ECG signals by sampling from a Markov chain starting from Gaussian noise, conditioned on noisy ECG inputs. It achieves state-of-the-art performance, outperforming baseline methods by at least 20% on four similarity metrics, with superior robustness under extreme noise corruption.

ABSTRACT

Objective: Electrocardiogram (ECG) signals commonly suffer noise interference, such as baseline wander. High-quality and high-fidelity reconstruction of the ECG signals is of great significance to diagnosing cardiovascular diseases. Therefore, this paper proposes a novel ECG baseline wander and noise removal technology. Methods: We extended the diffusion model in a conditional manner that was specific to the ECG signals, namely the Deep Score-Based Diffusion model for Electrocardiogram baseline wander and noise removal (DeScoD-ECG). Moreover, we deployed a multi-shots averaging strategy that improved signal reconstructions. We conducted the experiments on the QT Database and the MIT-BIH Noise Stress Test Database to verify the feasibility of the proposed method. Baseline methods are adopted for comparison, including traditional digital filter-based and deep learning-based methods. Results: The quantities evaluation results show that the proposed method obtained outstanding performance on four distance-based similarity metrics with at least 20\% overall improvement compared with the best baseline method. Conclusion: This paper demonstrates the state-of-the-art performance of the DeScoD-ECG for ECG baseline wander and noise removal, which has better approximations of the true data distribution and higher stability under extreme noise corruptions. Significance: This study is one of the first to extend the conditional diffusion-based generative model for ECG noise removal, and the DeScoD-ECG has the potential to be widely used in biomedical applications.

Motivation & Objective

• Address the critical need for high-fidelity ECG signal reconstruction in noisy clinical environments, especially during exercise stress tests where motion artifacts and baseline wander severely degrade signal quality.
• Overcome the limitations of traditional digital filters and deep learning-based methods, which often fail under high-amplitude noise or extreme corruption.
• Explore the potential of score-based diffusion models—previously unapplied to ECG denoising—for learning complex conditional distributions of clean ECG signals given noisy observations.
• Develop a method that better approximates the true data distribution and maintains stability even when noise levels are extreme.
• Enable more flexible and accurate ECG monitoring in real-world clinical and athletic settings by improving signal fidelity under challenging conditions.

Proposed method

• Propose a conditional score-based diffusion model tailored for ECG signals, learning the reverse denoising process through score matching on a Markov chain of noisy data distributions.
• Condition the diffusion process on the noisy ECG input at each step, enabling the model to generate clean ECG reconstructions by iteratively denoising from random Gaussian noise.
• Implement a multi-shot averaging strategy, where multiple independent reconstructions are generated and averaged to improve signal fidelity and reduce stochastic variation.
• Train the model using a denoising score matching objective that minimizes the difference between the predicted and true score functions of the clean ECG distribution.
• Leverage a U-Net-like architecture with residual blocks and attention mechanisms to model long-range dependencies in ECG time series, enhancing feature extraction and reconstruction quality.
• Optimize the model using a denoising score-based training objective that encourages the model to learn the score of the clean ECG distribution conditioned on the noisy input.

Experimental results

Research questions

• RQ1 Can a conditional score-based diffusion model achieve superior ECG denoising performance compared to traditional digital filters and deep learning-based methods?
• RQ2 How does the DeScoD-ECG model perform under extreme noise conditions, particularly in comparison to baseline methods that fail under high-amplitude corruption?
• RQ3 To what extent does the multi-shot averaging strategy improve reconstruction fidelity and reduce variance in the output signals?
• RQ4 Does the diffusion-based generative approach better approximate the true data distribution of clean ECG signals than end-to-end deep learning models?
• RQ5 Can the DeScoD-ECG model be efficiently deployed on standard personal computing hardware for real-world clinical use?

Key findings

• DeScoD-ECG achieved at least a 20% improvement in all four distance-based similarity metrics (e.g., NRMSE, SNR, etc.) compared to the best baseline method on the QT and MIT-BIH Noise Stress Test databases.
• The model demonstrated superior robustness under extreme noise corruption, maintaining high reconstruction fidelity where traditional and deep learning baselines failed.
• The multi-shot averaging strategy significantly enhanced reconstruction quality, reducing noise-induced artifacts and improving signal consistency.
• The diffusion-based approach better approximated the true data distribution of clean ECG signals, leading to more natural and physiologically plausible reconstructions.
• On a standard laptop (i7-10750H CPU, GTX 1660 Ti GPU), a 20-second ECG signal (7200 points) took 7.136 seconds for 10-shot reconstruction, indicating feasible but improvable inference speed.
• The model’s performance was validated across diverse ECG types and noise profiles, confirming its generalization potential for clinical and athletic monitoring applications.