BioTranslator: Inferring R-Peaks from Ambulatory Wrist-Worn PPG Signal.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Wearable sensors and wireless sensor networks (WSN) enable continuous biomedical signal monitoring.
• Wrist-worn photoplethysmography (PPG) sensors are popular for their form factor but suffer from motion artifacts in ambulatory settings, challenging vital parameter estimation.
• Accurate vital sign monitoring during physical activity is crucial for health and performance.

Purpose of the Study:

• To develop a novel deep learning framework, BioTranslator, for computing instantaneous heart rate (IHR) from wrist-worn PPG signals during physical activity.
• To address the challenge of signal degradation due to motion artifacts in wearable PPG sensors.
• To enable reliable IHR measurement in real-world, ambulatory conditions.

Main Methods:

• A novel deep learning framework, BioTranslator, utilizing a one-dimensional Convolution-Deconvolution Network.
• Translating single-channel PPG signals into an electrocardiogram (ECG)-like time series signal.
• Inferring R-peak information from the translated signal for IHR calculation.

Main Results:

• The BioTranslator network demonstrated high performance in identifying R-peaks from PPG signals (92.8% accuracy).
• Achieved a low mean absolute error of 51±6.3ms for instantaneous heart rate (IHR) compared to reference ECG-derived IHR.
• Evaluation conducted on 12 subjects from the TROIKA dataset during physical activity.

Conclusions:

• The BioTranslator framework offers a robust solution for accurate IHR estimation from wrist-worn PPG signals, even during physical activity.
• Deep learning translation of PPG to ECG-like signals effectively overcomes motion artifact challenges in wearable sensors.
• This advancement facilitates reliable, non-invasive vital parameter monitoring in ambulatory settings.