Non-Invasive Heart Rate Estimation From Ballistocardiograms Using Bidirectional LSTM Regression

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Machine Learning

Background:

• Non-invasive heart rate estimation is crucial for continuous cardiovascular disease monitoring.
• Ballistocardiogram (BCG) signals offer a promising, non-invasive method for heart rate assessment.
• Existing BCG-based methods face challenges like signal-to-reference mismatch and noise sensitivity.

Purpose of the Study:

• To develop a robust deep learning model for accurate non-invasive heart rate estimation from BCG signals.
• To address challenges in BCG signal processing, including sensor fusion and time-series feature learning.
• To investigate the impact of incorporating label uncertainty on heart rate estimation performance.

Main Methods:

• Development of a bidirectional long short-term memory (bi-LSTM) regression network.
• Utilizing BCG signals as input for the deep regression model.
• Incorporating label uncertainty into the estimation process to reduce annotation costs and improve performance.

Main Results:

• The proposed bi-LSTM network demonstrates strong fitting and generalization capabilities for BCG heart rate estimation.
• The model exhibits enhanced robustness against sensor noise and body movement perturbations.
• Performance improvements were observed when allowing for label uncertainty in the estimation.

Conclusions:

• The developed bi-LSTM regression network offers a reliable and effective solution for non-invasive heart rate estimation from BCG signals.
• This approach provides a more robust alternative for long-term cardiovascular health monitoring.
• The method effectively handles common challenges in BCG signal analysis, paving the way for improved clinical applications.