ResearchPod Summary
Accurate and long-term continuous blood pressure monitoring is critical for cardiovascular health management, yet traditional cuffless methods face severe limitations. Contact-based signals like photoplethysmography require sustained skin contact, while traditional one-dimensional ballistocardiography captures body recoil along only a single axis, missing energy leaked into other directions and remaining vulnerable to body-bed interaction variations. To overcome these challenges, the authors propose Phy-BP, a contactless blood pressure estimation framework based on triaxial bodyseismography, which measures bed vibrations across three orthogonal axes (X, Y, and Z).
The methodology consists of three core components. First, an adaptive quality-control algorithm utilizes a dual matched-filter approach—incorporating both a universal cardiogenic template and a dynamic template learned via linear time warping—to filter out low-quality, motion-corrupted segments. Second, the authors establish a physical model based on wave propagation theory in a viscoelastic medium to describe how a single cardiogenic excitation propagates through the body-bed system. Third, this physical model is embedded into an encoder-decoder deep learning architecture as a constraint, aligning multi-axis feature representations during training to improve robustness and generalization even under limited training samples.
The framework was evaluated using a comprehensive 162-hour hospital dataset collected from 21 subjects, employing continuous invasive arterial blood pressure measurements as the gold-standard reference alongside synchronized electrocardiography and photoplethysmography. Experimental results demonstrate that the proposed Phy-BP framework dynamically filters low-quality measurements and successfully leverages physical consistency across different axes. Consequently, the model provides faithful blood pressure monitoring that outperforms existing baseline methods and satisfies the stringent accuracy requirements established by the Association for the Advancement of Medical Instrumentation.
This study bridges the gap between raw mechanical bed vibrations and cardiovascular health monitoring by introducing a principled physical model to deep learning for physiological signal processing. By moving beyond traditional one-dimensional signals and black-box neural networks, Phy-BP offers a reliable, comfortable, and continuous alternative to cuff-based and invasive blood pressure monitoring, paving the way for practical overnight and long-term clinical care.
AI-generated third-party summary by ResearchPod. Not official content or an endorsement by the paper authors or affiliated organizations.