Robust graph-based denoising for cardiac acceleration signals.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiovascular Monitoring

Background:

• Heart failure monitoring relies on accurate physiological signals.
• Implantable devices offer continuous data but are susceptible to noise.
• Existing denoising methods may not fully capture the characteristics of heart vibration signals.

Purpose of the Study:

• To develop an efficient denoising approach for 3D accelerometer heart vibration signals.
• To improve the accuracy of heart failure monitoring using implantable devices.
• To leverage the pseudo-periodicity of cardiac signals for noise reduction.

Main Methods:

• Reformulating denoising as low-rank matrix inference.
• Utilizing graph-based signal smoothness assumptions.
• Applying group sparsity and total graph variation concepts.

Main Results:

• The proposed method demonstrated effectiveness in denoising real 3D accelerometer signals.
• Performance was validated on data from pigs with and without heart failure.
• The approach showed advantages over standard denoising techniques.

Conclusions:

• The novel denoising method enhances the quality of heart vibration signals.
• This technique holds promise for improving implantable device-based heart failure monitoring.
• Exploiting signal pseudo-periodicity and graph structures is a viable strategy for biomedical signal denoising.