Physics-informed machine learning enables continuous cuffless blood pressure monitoring

Blood pressure is a key metric of cardiovascular health, but standard methods for measuring it rely on occasional readings using inflatable cuffs, usually in a clinical setting. Today's blood pressure monitors are bulky, uncomfortable and only give readings while you're sitting still.

Now, an interdisciplinary team of mathematicians and engineers from the University of Utah and the University of Illinois, Chicago, is tackling this challenge by combining physics and artificial intelligence to overcome some of the limitations of existing devices. Appearing soon in Nature Communications, their study describes a new wearable smartwatch that can measure both blood pressure and blood flow continuously without needing a cuff.

"Elevated blood pressure is considered the silent killer because it leads to heart attacks, aneurysms and strokes. It represents a global healthcare burden and it is considered a Holy Grail problem," said Benjamin Sanchez Terrones, who hatched the project a few years ago as a Utahj assistant professor of electrical and computer engineering. It works by measuring the electrical properties of blood as it travels through the artery at the wrist, which fluctuate with changes in blood pressure.

The University of Utah holds the intellectual property associated with this technology, based on physics-informed machine learning, and the university's Technology Licensing Office is currently exploring licensing opportunities to bring this invention to market.

Light vs. electricity

The scientific basis of commercial wearable devices that use light to estimate blood pressure isn't fully understood, and often rely on machine learning as a "black box" to determine blood pressure, making their outputs difficult to interpret and clinically trust, the latter a major barrier for clinical adoption. Unlike these devices that measure light to gauge blood pressure, Sanchez Terrones' uses a painless and imperceptible electrical current.

The technology records tiny electrical changes in your wrist using bioimpedance, a measure of how easily electricity flows through blood and tissue. Because blood flow changes with each heartbeat, these electrical signals carry information about the underlying pressure.

This work shows how combining machine learning with physics can fundamentally change what's possible. By building physical principles directly into the model, we can move beyond black-box prediction toward systems that are more accurate, more interpretable, and more broadly applicable in real-world healthcare."

Christel Hohenegger, co-author, associate professor of mathematics

The roles of fluid dynamics and electromagnetism

The system harnesses fluid dynamics (how blood flows) and electromagnetism, giving it a clear scientific foundation and improving reliability. The model encodes the physics of pulsating blood and the electromagnetics of the bioimpedance measurement, so the network won't predict something that is physically impossible.

The result is a wearable device that can track cardiovascular health continuously, during rest and activity, without needing calibration to each individual user.

Utah graduate students Henry Crandall, Tyler Schuessler and Filip Bělík played a key role in testing the device on 150 actual people, including patients in intensive care and outpatient settings. "We went the extra mile and measured patients in the intensive care unit as well as the Madsen Health Center [a clinic just off campus in Salt Lake City] because we wanted to test the technology on the target population," said Sanchez Terrones, who last year relocated his lab to University of Illinois, Chicago, where he is an associate professor of electrical and computer engineering and biomedical engineering.

How blood and movies are alike

"Our blood pressure throughout the day is like a movie, but when you put on the cuff, all you get is one snapshot of the picture," said Sanchez Terrones. "The cuff device is very useful, but at the same time, limited: it only gives you the least amount of useful information because of the way the technology works: systolic readout over diastolic readout, which translates to the maximum and minimum pressure value during the recording. At the end, we are missing 99% of the movie that explains how blood pressure might change in a patient throughout the day while they are walking, running or climbing up stairs."

Sanchez Terrones' technology can catch the rest of the movie by recording velocity and pulse of blood as a continuous waveform, not just the familiar systolic and diastolic values provided in standard cuff readings like 120/80. (Systolic is the top number, measuring the pressure against the artery walls when the heart contracts, while diastolic is the pressure when the heart rests between beats.)

"Blood pressure isn't two numbers; it's a function of time. The mathematical challenge was recovering that whole waveform from indirect electrical measurements at the wrist-a classic inverse problem," said co-author Braxton Osting, a professor of mathematics at the University of Utah. "Embedding the physics of blood flow directly into the model makes the prediction more trustworthy."