Wearable ultrasound for continuous non-invasive blood pressure sensing

Stanford University, Stanford CA

Investigators

Abstract

Project Summary Hypertension is a leading risk factor for disease and death in the US, contributing to over 690,000 deaths annually. Blood pressure (BP) measurement is an essential part of a coordinated diagnosis and management strategy. There is a longstanding need for continuous non-invasive BP measurements to reveal the daily variations between clinic visits. Wearable devices such as smartwatches are attractive platforms for health sensing. However, existing wearable sensors fall short because they use a light-based signal called photoplethysmogram (PPG). While PPG biosignals are adequate for heart rate detection, their indistinct pulse wave shapes do not encode blood pressure. A promising alternative to PPG is wearable ultrasound, which can measure high resolution waveforms from large arteries. However, wearable ultrasound’s potential remains untested. The central hypothesis is that wearable ultrasound sensors can accurately estimate central blood pressure. This work aims to evaluate the central hypothesis using both simulated and experimental data. In Specific Aim 1, the BP estimation accuracy of wearable ultrasound will be evaluated using a large simulated dataset. An existing computational model of the arterial vasculature will be used to produce a dataset of simulated wearable ultrasound from the radial artery in the wrist. This dataset will be used to train convolutional neural networks to predict central BP. In Specific Aim 2, a prototype wearable ultrasound device will be developed and used to collect an experimental dataset with human subjects. 30 healthy subjects will wear wearable ultrasound while performing BP-altering activities including postural changes, physical activity, and controlled breathing. To convert measured recordings to a format congruent with the simulated data of Specific Aim 1, the experimental waveforms’ timing and shape information will be separated using a pulse deconvolution algorithm. A convolutional neural network will then be trained to predict brachial cuff pressure. To achieve these aims, the predoctoral fellowship applicant will conduct mentored research and training in cardiovascular simulations, wearable sensor design, and statistical algorithms for biomedical time series inference. The training plan includes coursework in cardiovascular simulations and statistical learning, hands-on lab skill training, regular advising meetings, and feedback on written and oral scientific exposition. In summary, this work combines simulated and experimental data to evaluate wearable ultrasound’s potential for BP estimation, opening a new avenue of research towards the longstanding need for continuous non-invasive BP sensing. Furthermore, through technical and professional skill development, the fellowship applicant will develop as an independent researcher advancing cardiovascular health through sensors and algorithms.