Heartbeat Detection and Rate Estimation from Ballistocardiograms using the Gated Recurrent Unit Network

Area of Science:

• Cardiovascular physiology
• Biomedical signal processing
• Machine learning in healthcare

Background:

• Heartbeat detection is crucial for cardiovascular health monitoring.
• Traditional methods face challenges in accurately estimating heart rate from ballistocardiogram (BCG) signals.
• Recurrent Neural Networks (RNNs) show promise in signal processing applications.

Purpose of the Study:

• To propose a novel heartbeat detection framework using Gated Recurrent Unit (GRU) networks.
• To model BCG signal segments as images for GRU network input.
• To enhance heart rate estimation accuracy by leveraging multi-channel BCG signals and extracting temporal/waveform characteristics.

Main Methods:

• Developed a heartbeat detection framework utilizing a Gated Recurrent Unit (GRU) network.
• Formulated BCG signal segments as images for classification.
• Employed GRU for feature extraction from BCG signal images.
• Integrated multi-channel BCG signals for improved performance.

Main Results:

• The proposed GRU-based framework achieved superior heart rate estimation results compared to existing algorithms.
• Demonstrated effective fusion of multi-channel BCG signals.
• Showcased effective extraction of temporal and waveform characteristics of heartbeat signals.

Conclusions:

• The GRU network offers an effective approach for heartbeat detection from BCG signals.
• The proposed framework significantly enhances heart rate estimation accuracy.
• This method holds potential for improved non-invasive cardiovascular monitoring.