Contactless Cardiac Health Monitoring with Millimeter-Wave Radar Based on PMG-SATNet.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Cardiology

Background:

• Cardiovascular diseases (CVDs) are leading causes of global mortality, often presenting subtly.
• Traditional electrocardiogram (ECG) monitoring uses skin electrodes, which can cause irritation and limit routine use.
• Non-contact cardiac monitoring using millimeter-wave radar and deep learning is an emerging research area.

Purpose of the Study:

• To develop a robust deep learning model for high-fidelity ECG signal recovery from millimeter-wave radar data.
• To address the challenge of poor generalization in single-source dataset training by creating diverse experimental scenarios.
• To improve the accuracy and reliability of non-contact cardiac monitoring for early risk assessment.

Main Methods:

• A novel deep learning network, PMG-SATNet, featuring encoder-decoder structures was designed.
• The encoder utilizes parallel multi-scale feature extraction and global temporal modeling for comprehensive pattern capture.
• The decoder incorporates a spectral attention-augmented temporal convolutional network to filter noise and highlight relevant ECG frequencies.

Main Results:

• PMG-SATNet demonstrated superior performance compared to baseline models on a self-built dataset.
• Significant improvements were observed in Pearson correlation coefficient (3.3% and 3.8%) and root mean square error (16.4% and 23.8%).
• The model effectively recovered ECG signals from radar-derived chest vibrations with high fidelity.

Conclusions:

• PMG-SATNet shows high fidelity in recovering ECG signals from millimeter-wave radar data.
• The proposed method offers a potential solution for real-life, non-contact cardiac health monitoring.
• This technology could enhance early detection and management of cardiovascular diseases.