Leveraging conditional diffusion and pruning for optimal ECG signal denoising

Electrocardiograms (ECGs) are essential for diagnosing a wide range of cardiac conditions, but their accuracy is often compromised by noise and baseline wander, particularly in real-world settings. Traditional denoising techniques struggle to preserve the intricate features of ECG signals while removing noise, whereas advanced methods often suffer from higher computational costs, limiting their practical use. To address these challenges, this paper proposes an innovative solution using the Improved Denoising Diffusion Probabilistic Model (IDPM) with a novel combination of a conditional framework and a Quality Assignment Pruning technique. This approach leverages the exceptional ability of diffusion models to process complex datasets and their proven effectiveness in denoising tasks, achieving both improved performance and reduced computational cost. Our approach selectively enhances the most relevant features of ECG signals, enabling precise noise reduction and baseline correction. Experimental results demonstrate that our method significantly outperforms state-of-the-art advanced and traditional models in ECG denoising, even under extreme noise conditions. With reduced computational requirements suitable for low resource clinical devises, this approach offers a robust and efficient solution that opens up more research directions for other biomedical signal applications.