Hyde: Inferring Cardiovascular Hemodynamics Using BCG for Cross-User Hypertension Detection

Hypertension, a leading contributor to global mortality, affects 1.28 billion adults worldwide, and its early detection is essential to prevent adverse outcomes. However, relying solely on blood pressure for hypertension detection proves insufficiently accurate. Recognizing this limitation, we have identified a range of multi-dimensional hemodynamics as novel biomarkers that can offer more comprehensive cardiovascular insights for detecting hypertension. Specifically, we propose Hyde, a cross-user hypertension detection system that leverages ballistocardiogram (BCG) signals to infer multi-dimensional hemodynamics and naturally detect hypertension without disrupting users' daily routines. Hyde features a multi-task, multi-branch, unsupervised domain adaptation learning framework, enabling simultaneous prediction of five hemodynamic biomarkers in the auxiliary regression task. This design addresses several challenges, including unlabeled target domain, single-source bias, and complete negative transfer. We implement a flexible optic fiber sensor mat to collect high-quality BCG signals and deploy it in a four-month clinical trial involving 85 subjects with different conditions. The results show that Hyde can accurately identify hypertension with an average accuracy of 97.65%, outperforming state-of-the-art approaches by 21.65% \sim 38.16%. To further validate Hyde's reliability, we test the system with another 33 participants under natural, unconstrained sleep conditions, revealing consistent performance. These findings highlight Hyde's potential for long-term, non-contact hypertension assessment, offering valuable clinical implications for personalized healthcare and improving cardiovascular outcomes.