Repaint High-Density Surface Electromyography Signal Using Denoising Diffusion Probabilistic Model.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• High-density surface electromyography (HD-sEMG) is crucial for myoelectric control and muscle activation analysis.
• Signal corruption and loss due to poor electrode contact hinder the practical application of HD-sEMG.
• Existing interpolation methods are insufficient for reconstructing complex, multi-channel corrupted signals.

Purpose of the Study:

• To develop a novel and effective approach for reconstructing corrupted HD-sEMG signals.
• To overcome the limitations of conventional interpolation methods in handling multi-channel signal loss.
• To enhance the fidelity and reliability of HD-sEMG data acquisition.

Main Methods:

• A denoising diffusion probabilistic model (DDPM) with a repaint strategy was employed for HD-sEMG signal reconstruction.
• A U-Net architecture incorporating spatiotemporal embedding modules was utilized to capture signal characteristics.
• The method reconstructs signals without requiring prior knowledge of the corruption patterns.

Main Results:

• The proposed DDPM approach significantly outperformed linear/cubic interpolation, GAN, and VAE methods in signal reconstruction accuracy (lower nRMSE).
• Achieved the lowest average normalized root-mean-square error (nRMSE) of 0.027$\pm$0.027 across various corruption ratios.
• Demonstrated superior performance in peak signal-to-noise ratio (PSNR) and maintained robust classification accuracy, comparable to ground truth.

Conclusions:

• The novel DDPM-based reconstruction method offers a significant advancement in HD-sEMG signal processing.
• This approach enhances the fidelity and reliability of HD-sEMG signals, enabling more robust myoelectric control.
• Provides a promising solution for real-world applications where signal integrity is critical.