Study on Multimodal Sensor Fusion for Heart Rate Estimation Using BCG and PPG Signals

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiovascular Health

Background:

• Continuous heart rate monitoring is vital for early cardiovascular disease detection.
• Electrocardiography (ECG) presents limitations for comfortable, long-term home monitoring.
• Non-invasive physiological signals offer potential for remote health assessment.

Purpose of the Study:

• To develop a multimodal temporal fusion network (MM-TFNet) for accurate heart rate monitoring.
• To integrate ballistocardiography (BCG) and photoplethysmography (PPG) signals for enhanced cardiovascular monitoring.
• To overcome the discomfort and limitations associated with traditional ECG monitoring in home environments.

Main Methods:

• Utilized temporal convolutional networks (TCNs) and bidirectional long short-term memory networks (BiLSTMs) to extract temporal features from BCG and PPG signals.
• Implemented a cross-modal attention mechanism for adaptive fusion of BCG and PPG features.
• Employed multi-head self-attention (MHSA) to focus on key heartbeat waveforms, improving robustness during dynamic activities.

Main Results:

• The MM-TFNet achieved a mean absolute error (MAE) of 0.88 BPM in heart rate prediction.
• The proposed model outperformed existing mainstream deep learning methods in heart rate prediction tasks.
• Experimental validation was conducted on a public BCG-PPG-ECG dataset with 40 subjects.

Conclusions:

• The MM-TFNet demonstrates high accuracy and robustness for continuous heart rate monitoring using non-invasive BCG and PPG signals.
• This multimodal fusion approach provides a foundation for contactless, low-power, edge-deployable home health monitoring systems.
• The study highlights the significant potential of multimodal fusion in analyzing complex physiological signals for improved cardiovascular health management.