Non-Contact Blood Pressure Estimation From Radar Signals by a Stacked Deformable Convolution Network

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Traditional blood pressure monitoring is invasive and can be uncomfortable.
• Contactless methods are needed for continuous and convenient physiological monitoring.
• Radar technology offers a non-invasive way to capture subtle physiological signals.

Purpose of the Study:

• To develop and validate a contactless method for continuous blood pressure estimation.
• To leverage deep learning for enhanced feature extraction from radar signals.
• To assess the performance of a novel deep learning architecture for blood pressure monitoring.

Main Methods:

• Utilized a 24 GHz continuous wave radar system to collect chest micro-variation data.
• Preprocessed radar signals using Kalman filtering, multiscale bandpass filters, and periodic extraction.
• Developed the Radar-based Stacked Deformable convolution Network (RSD-Net) with attention mechanisms for feature extraction and blood pressure estimation.

Main Results:

• Achieved high accuracy with Pearson correlation coefficients of 0.838 for systolic and 0.797 for diastolic blood pressure.
• Reported low Mean Error (ME) and Standard Deviation (SD): -0.32 ±6.14 mmHg (systolic) and -0.20 ±5.50 mmHg (diastolic).
• Ablation studies confirmed the significance of individual components within the RSD-Net architecture.

Conclusions:

• The proposed contactless radar-based system with RSD-Net effectively estimates continuous blood pressure.
• The method demonstrates robustness across various physiological conditions (rest, Valsalva, apnea, tilt-table).
• This technology holds promise for non-invasive, long-term blood pressure monitoring.