Wearable ultrasonic sensor monitors people on the move

Engineers in the US have developed a fully integrated wearable ultrasound system for deep tissue monitoring that works when subjects are in motion.

A wearable ultrasonic-system-on-patch for deep tissue monitoring
A wearable ultrasonic-system-on-patch for deep tissue monitoring - Muyang Lin

The work by a team from the University of California San Diego is claimed to facilitate potentially life-saving cardiovascular monitoring. Their work is detailed in the May 22, 2023 issue of Nature Biotechnology.

“This project gives a complete solution to wearable ultrasound technology—not only the wearable sensor, but also the control electronics are made in wearable form factors,” said Muyang Lin, a Ph.D. candidate in the Department of Nanoengineering at UC San Diego and the first author on the study. “We made a truly wearable device that can sense deep tissue vital signs wirelessly.”

The research is from the lab of Sheng Xu, a professor of nanoengineering at UC San Diego Jacobs School of Engineering and corresponding author of the study. 

This fully integrated autonomous wearable ultrasonic system-on-patch (USoP) is said to build on the lab’s previous work in soft ultrasonic sensor design. These soft ultrasonic sensors require tethering cables for data and power transmission, which largely constrains the user’s mobility. This work includes a small, flexible control circuit that communicates with an ultrasound transducer array to collect and transmit data wirelessly. A machine learning component helps interpret the data and track subjects in motion. 

According to the lab’s findings, the ultrasonic system-on-patch allows continuous tracking of physiological signals from tissues as deep as 164mm, continuously measuring central blood pressure, heart rate, cardiac output, and other physiological signals for up to twelve hours at a time.

“This technology has lots of potential to save and improve lives,” Lin said in a statement. “The sensor can evaluate cardiovascular function in motion. Abnormal values of blood pressure and cardiac output, at rest or during exercise, are hallmarks of heart failure. For healthy populations, our device can measure cardiovascular responses to exercise in real time and thus provide insights into the actual workout intensity exerted by each person, which can guide the formulation of personalised training plans.”

According to the team, the USoP also represents a breakthrough in the development of the Internet of Medical Things (IoMT), a term for a network of medical devices connected to the internet, wirelessly transmitting physiological signals into the cloud for computing, analysis and professional diagnosis.

“At the very beginning of this project, we aimed to build a wireless blood pressure sensor,” said Lin. “Later on, as we were making the circuit, designing the algorithm and collecting clinical insights, we figured that this system could measure many more critical physiological parameters than blood pressure, such as cardiac output, arterial stiffness, expiratory volume and more, all of which are essential parameters for daily health care or in-hospital monitoring.”

Moreover, when the subject is in motion, there will be relative movement between the wearable ultrasonic sensor and the tissue target, which will require frequent manual readjustment of the wearable ultrasonic sensor to keep track of the moving target. In this work, the team developed a machine learning algorithm to automatically analyse the received signals and choose the most appropriate channel to keep track of the moving target. 

When the algorithm is trained using one subject’s data, that learning may not be transferable to other subjects, making the results inconsistent and unreliable.

“We eventually made the machine learning model generalisation work by applying an advanced adaptation algorithm,” said Ziyang Zhang, a master’s student in the Department of Computer Science and Engineering at UC San Diego and co-first author on the paper. “This algorithm can automatically minimise the domain distribution discrepancies between different subjects, which means the machine intelligence can be transferred from subject to subject. We can train the algorithm on one subject and apply it to many other new subjects with minimal retraining.”

The sensor, set for tests among larger populations, will be commercialised by Softsonics, LLC.