Efficient J Peak Detection From Ballistocardiogram Using Lightweight Convolutional Neural Network.

Area of Science:

• Biomedical Engineering
• Cardiovascular Monitoring
• Artificial Intelligence in Healthcare

Background:

• Ballistocardiagram (BCG) offers non-contact, non-invasive physiological monitoring for cardiovascular disease (CVD).
• Accurate J-peak detection in BCG signals is crucial for deriving key cardiovascular indicators.
• Existing deep learning models (CNN, RNN) for J-peak detection face challenges with inference speed and complexity.

Purpose of the Study:

• To develop a computationally efficient and memory-optimized neural network for J-peak detection in BCG signals.
• To enhance J-peak detection performance through a novel physiological meaning-based transformation method.
• To evaluate the proposed model's accuracy, latency, and size against baseline methods.

Main Methods:

• Proposed JwaveNet, a robust lightweight neural network model for J-peak detection.
• Implemented a new transformation method for J-peaks based on physiological meaning during preprocessing.
• Collected synchronous BCG and electrocardiogram (ECG) data from 24 subjects across four sleeping positions.

Main Results:

• JwaveNet demonstrated significant reductions in latency and model size compared to existing baseline models.
• The novel J-peak transformation method improved detection performance.
• The lightweight model achieved high accuracy in J-peak detection from BCG signals.

Conclusions:

• JwaveNet offers a computationally efficient solution for J-peak detection in BCG signals.
• The proposed method enhances the feasibility of using BCG for remote CVD monitoring.
• This lightweight deep learning approach holds promise for improving home-based cardiovascular health assessment.