If you own a sphygmomanometer—the gadget with the inflatable cuff that’s used to measure blood pressure—and if you’re diligent about using it, you might take your own blood pressure as often as a few times a day. Otherwise, you probably get your blood pressure checked only when you visit a doctor, perhaps a few times a year.
Is that enough? Blood pressure varies over time, not just year to year or day to day, but sometimes minute to minute. It rises when we’re active or feeling stressed, and it falls when we’re relaxed. Individuals with so-called white-coat hypertension, who feel unduly nervous around doctors, might get their blood pressure checked only when it’s unnaturally high—and thus end up saddled with medications or treatments they don’t really need.
In a world of smart watches and fitness monitors that continuously monitor heart rhythm, skin temperature, sleep quality, and more, blood pressure stands out as a key quantity that’s absent from the devices’ suite of measurements. The sphygmomanometer is just too big and inconvenient.
To fill the blood-pressure data gap, Roozbeh Jafari (Texas A&M University), Deji Akinwande (the University of Texas at Austin), and their colleagues have developed a new blood-pressure sensor that’s light and unobtrusive enough to be carried around everywhere.1 The sensor uses a temporary tattoo made of graphene and protected by an ultrathin polymer film (the shiny patches in figure 1), and it works by measuring bioimpedance—essentially the tissue’s resistance to an alternating electrical current—as blood pulses through the artery under the tattoo.
Six graphene electrodes (the shiny patches), lined up over the left radial artery, measure bioimpedance in the wrist. Through a machine-learning algorithm, the bioimpedance data are converted into a blood-pressure measurement. (Courtesy of Roozbeh Jafari and Deji Akinwande.)
Six graphene electrodes (the shiny patches), lined up over the left radial artery, measure bioimpedance in the wrist. Through a machine-learning algorithm, the bioimpedance data are converted into a blood-pressure measurement. (Courtesy of Roozbeh Jafari and Deji Akinwande.)
At the moment, converting bioimpedance to blood pressure is a task for a complicated machine-learning algorithm that requires hours of training on each new user. The researchers hope to develop their sensor into a plug-and-play device that’s just as portable as a smart watch, so they can add blood pressure to the set of continuously monitored vital signs.
The body electric
The sphygmomanometer is a century-old technology. Its mechanism of operation hasn’t changed much since Russian surgeon Nikolai Korotkov recognized, in 1905, the important distinction between systolic blood pressure (the highest pressure at the peak of a blood pulse) and diastolic blood pressure (the lowest pressure between blood pulses). To start, the cuff inflates until it squeezes an artery completely shut. Then the cuff gradually deflates until blood can push its way through on part, then all, of the heartbeat cycle. The result is a pair of pressures, measured in millimeters of mercury, that make up the standard blood-pressure measurement.
The need to squeeze hard enough to cut off blood flow explains why the sphygmomanometer cuff is so bulky and why its operation is so uncomfortable. Some emerging technologies can measure blood pressure using smaller, gentler pressure sensors.2 But Jafari, Akinwande, and colleagues sought to create a blood-pressure gauge that doesn’t measure mechanical pressure at all, but rather other quantities that are correlated with it.
Jafari, an expert on machine-learning algorithms for biomedical applications, has been working for several years on estimating blood pressure from bioimpedance.3 Blood is a fluid rich in ions, so it’s a better electrical conductor than most other tissues are. When a blood pulse passes through an artery, the overall tissue impedance drops, as shown in figure 2. Furthermore, higher blood pressure is correlated with faster propagation of blood pulses. So if bioimpedance is measured at two spots on the same artery, the pulse transit time between them is another valuable piece of information.
Blood pulsing beneath the electrodes shown in figure 1 changes the tissue’s electrical conductivity, and thus its impedance. As sketched here, the blood-pressure and bioimpedance curves are inversely related to each other. To extract an absolute blood-pressure measurement, a machine-learning algorithm is fed data, such as the impedance peaks and troughs (yellow triangles) and the pulse transit time between two pairs of electrodes on the same artery. (Adapted from ref. 1.)
Blood pulsing beneath the electrodes shown in figure 1 changes the tissue’s electrical conductivity, and thus its impedance. As sketched here, the blood-pressure and bioimpedance curves are inversely related to each other. To extract an absolute blood-pressure measurement, a machine-learning algorithm is fed data, such as the impedance peaks and troughs (yellow triangles) and the pulse transit time between two pairs of electrodes on the same artery. (Adapted from ref. 1.)
Still, blood pressure and impedance aren’t precisely linked by any simple, direct relationship. The blood-pressure measurement needs to be teased out from subtle features of the shape of the impedance curve over the pulse cycle. That’s where machine learning comes in.
At first, Jafari and colleagues measured bioimpedance using conventional electrodes similar to those used to record electrocardiograms. The electrodes had to be held in place with a strong adhesive, so they could be uncomfortable to wear for a long time. Worse, as the adhesive layer deformed over time, the electrodes would shift position on the skin, which introduced measurement error. That’s not practical for a device meant to collect measurements continuously.
Eye, robot
Meanwhile, Akinwande was exploring applications for biomedical sensors made from graphene. Ever since the two-dimensional carbon material burst onto the scientific scene almost two decades ago, researchers have been devising graphene-based technologies and experiments that previously would have been almost unimaginable. Says Akinwande, “I’ve been asking myself, what opportunities are there that only graphene can provide?”
Many of those opportunities have come in the area of flexible, wearable electronics. Graphene is a semimetal, so it can be fashioned into electrodes to pick up electrical signals from the human body or anywhere else. It’s atomically thin, and it clings to the skin by van der Waals forces alone, with no adhesive required.
Even with a protective polymer layer on top, needed to keep the graphene from rubbing right off, the whole structure is only a fraction of a micron thick. (Conventional electronics can attach to the skin through van der Waals forces too—see Physics Today, May 2019, page 16—but they’re orders of magnitude thicker.) Graphene is thin and flexible enough to conform to all the wrinkles and corrugations of the skin without the wearer even noticing it’s there.
Akinwande calls the electrodes “graphene electronic tattoos” because of their similarity to ink-based temporary tattoos: The graphene and polymer are layered onto a specially formulated tattoo paper, which is pressed against the skin and wetted with water to transfer the tattoo into place. Surprisingly rugged, the tattoos aren’t damaged by daily activities or even gentle washing, and they can remain in place for up to a week before inevitably getting sloughed off with a layer of dead skin cells.
The tattoos’ applications are manifold. In one of their experiments, Akinwande and colleagues placed the graphene tattoos on the skin around a subject’s eyes to measure the electrical signals from the eye muscles and determine, with few-degree precision, where the subject was looking. They used the signals to pilot a robotic drone, which the subject could steer just by looking around the room.4
For blood-pressure measurements, the important thing about the tattoos is that they don’t shift in position over time. Their steady measurements are ideal for training, then using, Jafari’s machine-learning algorithm.
Continuous learning
As shown in figure 1, the new blood-pressure sensor uses six graphene patches lined up over the radial artery, on the side of the wrist nearest the thumb. Six more, not shown, cover the ulnar artery on the other side. In each set, the electrodes on each end inject an imperceptibly tiny electric current into the wrist. The other four are split into two pairs, each of which measures the induced potential difference, which is proportional to the impedance.
To train the machine-learning algorithm, the researchers had a handful of volunteers wear both the graphene tattoos and conventional sphygmomanometers for several hours while they performed activities designed to raise and lower their blood pressure. The researchers fed the machine-learning algorithm with several quantities extracted from the bioimpedance curves: the peak and trough of each pulse cycle, the maximum slope, and the pulse transit time, among others. By the end of the training, the algorithm was estimating both systolic and diastolic blood pressure to within about 5 mm Hg—an excellent result, as blood-pressure measurements go.
Obviously, most of the advantage of a lightweight blood-pressure gauge is negated if you have to wear a sphygmomanometer for hours to train it. Ideally, the researchers would like to be able to apply a graphene tattoo to a new subject and immediately get accurate blood-pressure readings. They’re not there yet, but they hope to make progress by exploring different quantities from the bioimpedance curves—ratios rather than absolute values, for example—that might be less user-specific. And in a promising result, several days after the experiment they applied a new tattoo to one of the study participants. With no additional algorithm training, they got blood-pressure readings precise to within 10 mm Hg: not as good as in the original experiment, but still a useful measurement.
Jafari suspects that if continuous blood-pressure monitoring becomes standard, it could lead to a shift in thinking about blood-pressure measurement errors. Under the current paradigm, readings are taken infrequently, so each measurement and its error bars are treated in isolation. “But with continuous monitoring, the absolute error is less important,” he says. “What’s important is the trend. Does your blood pressure drop when you go to bed, and does it increase when you’re stressed out? By how much?”
Another advantage of the tattoos is that they can be applied nearly anywhere on the body. A sphygmomanometer cuff is almost always wrapped around the brachial artery in the upper arm, and although the resulting measurement is presented as “your blood pressure,” blood pressure is affected by local conditions, such as the stiffness of individual arteries, so it’s not the same everywhere in the body. Patients with poor blood circulation, especially, can potentially benefit from data on blood pressure in different parts of the body—such as arteries in the neck that supply blood to the brain—that are too dangerous or impractical to probe with the century-old artery-squeezing technology.
