Self-Supervised, Non-Contact Heartbeat Detection Based on Ballistocardiograms Utilizing Physiological Information...

Area of Science:

• Biomedical Engineering
• Physiological Signal Processing
• Machine Learning in Healthcare

Background:

• Ballistocardiograms (BCG) offer passive, non-contact heart rate monitoring but are susceptible to external disturbances and signal degradation.
• The prominence of the j-peak in BCG signals diminishes with declining health, impacting accuracy.
• Existing methods struggle with BCG signal robustness in diverse and adverse conditions.

Purpose of the Study:

• To develop a non-contact, self-supervised heart rate detection method using BCG signals.
• To enhance the accuracy and robustness of BCG-based heart rate detection under challenging conditions.
• To improve the physiological significance and reliability of BCG signal analysis.

Main Methods:

• A self-supervised algorithm guided by BCG heart rate estimation to reconstruct physiologically significant signals.
• A Bidirectional Long Short-Term Memory Network (BiLSTM) based heartbeat mapping algorithm for deep feature extraction and real-time prediction.
• Evaluation of the method on 40 young and 4 elderly subjects.

Main Results:

• The proposed method demonstrated excellent performance in beat-to-beat heart rate estimation and heartbeat detection.
• It surpassed existing state-of-the-art methods, including those using precise labels.
• Effective heartbeat detection was achieved, showing robustness against noise and variations.

Conclusions:

• The developed non-contact, self-supervised BCG method significantly improves heart rate detection accuracy and robustness.
• The BiLSTM-based algorithm effectively extracts features and predicts heartbeats, mitigating reconstruction deviations.
• This approach offers a reliable solution for physiological monitoring in various conditions.