Learning based Quality Indicator Aiding Heart Rate Estimation in Wrist-Worn PPG.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Signal Processing

Background:

• Wearable sensors enable continuous physiological monitoring, but wrist-worn photoplethysmography (PPG) devices struggle with motion artifacts during daily activities.
• Poor signal quality from PPG sensors hinders accurate estimation of vital cardiac parameters.

Purpose of the Study:

• To develop a learning-based Quality Indicator Engine (QIE) to assess the reliability of wrist-worn PPG signals.
• To evaluate the QIE's effectiveness in improving heart rate (HR) estimation in ambulatory settings.

Main Methods:

• The QIE utilizes frequency-domain feature extraction, feature selection, and an ensemble of decision trees for classification.
• The engine was evaluated on 23 PPG records from the TROIKA database, focusing on data acquired during physical activities.

Main Results:

• The QIE achieved 83% accuracy in classifying PPG signal quality on the testing set.
• Integration of the QIE with a state-of-the-art WFPV algorithm resulted in a 43% average improvement in heart rate estimation.

Conclusions:

• The developed QIE is the first to be evaluated on wrist-PPG data during physical activities for improved HR estimation.
• The QIE enhances the efficacy of vital parameter estimation from wrist-worn PPG sensors in real-world, ambulatory conditions by mitigating motion artifact interference.