Reduction of motion artifacts from photoplethysmography signals using learned convolutional sparse coding.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Wearable photoplethysmography (PPG) offers continuous, non-invasive cardiac monitoring for cardiovascular disease management.
• Motion artifacts in daily life severely degrade PPG signal quality, challenging traditional denoising methods.
• Current deep learning denoisers lack interpretability, hindering clinical adoption.

Purpose of the Study:

• To develop a novel framework combining signal decomposition and deep learning for interpretable PPG signal denoising.
• To enhance the robustness of wearable PPG monitoring against motion artifacts.

Main Methods:

• An algorithm unfolding approach integrated prior PPG knowledge into a deep neural network for improved interpretability.
• A learned convolutional sparse coding model was employed for signal representation.
• The network was trained using the PulseDB dataset and a synthetic artifact model, then validated on the PPG-DaLiA dataset.

Main Results:

• The proposed method improved signal-to-noise ratio (SNR) by 18.29 dB on a synthetic dataset.
• Heart rate mean absolute error (MAE) was reduced by 55% on synthetic data and 23% on real-world data.
• The framework achieved higher SNR and comparable MAE to existing deep learning methods.

Conclusions:

• The developed method effectively enhances PPG signal quality from wearable devices.
• It enables extraction of meaningful waveform features for improved cardiovascular disease monitoring.
• This approach holds potential for developing innovative tools for remote cardiac health assessment.