Learning to Estimate Heart Rate From Accelerometer and User's Demographics During Physical Exercises.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Signal Processing

Background:

• Wearable devices offer non-invasive health insights, with heart rate (HR) monitoring being crucial.
• Photoplethysmography (PPG) is common for real-time HR estimation but is susceptible to motion artifacts (MA).
• Existing methods struggle with HR estimation during high-intensity exercises like running due to MA.

Purpose of the Study:

• To develop a novel HR estimation method for wearables that overcomes PPG limitations during motion.
• To enhance HR prediction accuracy using accelerometer signals and user demographics when PPG is compromised.
• To enable on-device personalization and HR prediction even without PPG signals.

Main Methods:

• A new algorithm integrating accelerometer data and user demographics for HR prediction.
• On-device personalization through real-time finetuning of model parameters during workouts.
• Evaluation on diverse exercise datasets, including treadmill and outdoor activities.

Main Results:

• The proposed method improves the coverage of PPG-based HR estimators.
• Maintained similar error performance compared to existing methods.
• Demonstrated utility in predicting HR for short durations without PPG data.

Conclusions:

• The new algorithm effectively supports HR prediction during motion artifacts, enhancing wearable device functionality.
• On-device personalization and the ability to predict HR without PPG offer significant improvements for user experience.
• This method provides a robust solution for accurate HR monitoring during various physical activities.