Contactless and continuous blood pressure detection with FMCW millimeter-wave radar

Blood pressure (BP) is a vital sign that could provide crucial information for the prevention, diagnosis and treatment of many diseases. Most of the existing BP detection solutions require physical contact between human skin and sensor, which may be frustrated due to uncomfortable experience and extra burden. Recently, millimeter-wave (mm-wave) radar based BP detection has attracted more attention, but it is confronted with the challenges on noise interference, fine-grained arterial pulse signal recovery and shortage of annotated samples for training BP estimation model. In this paper, we develop mmRBP, a contactless and continuous BP detection system with a single frequency modulated continuous wave (FMCW) mm-wave radar. mmRBP consists of three modules. The arterial pulse motion extraction module is designed to mitigate the noise interference, in which the range fast Fourier transform (range-FFT) and digital beamforming (DBF) are applied to scan the space and the direction with the strongest arterial pulse motion is identified. Then, the arterial pulse signal translation module is devised to transform the mm-wave reflections to fine-grained arterial pulse signal by the convolutional neural network (CNN) based translation filter. Finally, the BP estimation module is designed to address the issue of shortage of annotated radar training samples, where a deep transfer learning (DTL) model is employed to transfer the knowledge of the source network trained by the public BP dataset to the target network trained by the collected radar dataset. Results of extensive experiments show that the error of mmRBP is 0.87±6.12 mmHg (Mean error±Standard Deviation) and 0.59±3.78 mmHg for systolic BP (SBP) and diastolic BP (DBP), which satisfy both the British Hypertension Society (BHS) and the Association for the Advancement of Medical Instrumentation (AAMI) BP measurement standards.