KAN-DeScoD: Kolmogorov-Arnold Network Enhanced Deep Score-Based Diffusion Model for ECG Denoising

Area of Science:

• Biomedical Signal Processing
• Machine Learning
• Artificial Intelligence

Background:

• Deep score-based diffusion (DeScoD) models show promise for electrocardiogram (ECG) denoising.
• However, limitations exist in handling non-stationary noise and complex ECG features due to linear transformation approximations.

Purpose of the Study:

• To introduce the Kolmogorov-Arnold network enhanced deep score-based diffusion (KAN-DeScoD) model for improved ECG denoising.
• To enhance the flexibility and accuracy of ECG signal reconstruction, particularly in high-noise environments.

Main Methods:

• Integration of Kolmogorov-Arnold network (KAN) layers into a deep score-based diffusion model.
• Leveraging KAN's adaptive activation functions to capture intricate ECG signal structures.
• Validation on the QT Database and MIT-BIH Noise Stress Test Database (NSTDB).

Main Results:

• The KAN-DeScoD model demonstrated superior performance compared to the DeScoD model across various metrics.
• Improved robustness in high-noise environments and enhanced accuracy and stability in signal reconstruction were observed.
• Effectiveness validated under different noise intensities and sampling rates.

Conclusions:

• The integration of KAN layers significantly enhances the performance of diffusion models for ECG denoising.
• KAN-DeScoD offers a more robust and accurate solution for reconstructing complex ECG signals, especially under challenging noise conditions.