PracticalBP: Continuous Cuffless Blood Pressure Monitoring with Only One Record for Calibration

Frequent blood pressure (BP) monitoring is crucial for helping patients and clinicians assess the risk of cardiovascular disease. The current oscillometric method needs to occlude peripheral arteries and the repeated inflation can cause discomfort, leading to reduced compliance. Although several cuffless BP monitoring systems have been proposed, they often either lack sufficient accuracy or impose a significant calibration burden, rendering them impractical. To address this gap, we introduce PracticalBP, a cuffless wrist-worn system that requires only a single BP reading for calibration to achieve accurate BP measurements. We identify the main issue of current cuffless devices is that the key user-specific hemodynamic information cannot be obtained with a low effort. Our key insight is that these individual BP-related information can be obtained effortlessly by physiology-based methods. To facilitate this, we propose a physiology-based pseudo-label generation mechanism to create an individual data distribution based on the principles of tonometry and hydrostatic pressure in arteries. During it, the user only needs to simply stay calm and tilt the arm. Furthermore, we observe that BP-related features exhibit different patterns across individuals. Therefore, we propose an effective feature extraction approach that combines a general feature extractor with a feature-wise attention module to assess the importance of each feature for different subjects. Additionally, in order to improve robustness, we design a channel-based data augmentation mechanism to cope with the signal distortion caused by the sensor position changes during each wear. PracticalBP is implemented in the real system with our developed wristband. We collect a dataset from 25 healthy subjects and 18 hypertension patients and conduct both rest tests and BP intervention experiments. The evaluation demonstrates that PracticalBP achieves an overall systolic BP (SBP) and diastolic BP (DBP) estimated error (mean ± standard deviation) of (-1.49 ± 7.89) mmHg and (-0.83 ± 6.23) mmHg, respectively, which meets the requirement of AAMI standard. Notably, our system achieves comparable accuracy to state-of-the-art solutions while reducing the calibration time by a factor of five and eliminating the need for complex BP intervention tasks.