A 30-second selfie video can now screen for hypertension and diabetes

Checking your blood pressure means strapping on an uncomfortable arm cuff, and screening for diabetes usually involves sticking a needle into your finger. For millions of people, that friction alone is enough to avoid testing altogether.

Researchers at the University of Tokyo are trying to replace all of that with a glance. In a preliminary study presented at the American Heart Association's Scientific Sessions 2024, the team unveiled a system that flags high blood pressure and diabetes simply by filming a person's face and palms for a few seconds.

Reading blood flow at 150 frames per second

Led by project researcher Ryoko Uchida, the system relies on high-speed video recording at 150 images per second. Because both hypertension and diabetes cause subtle microvascular shifts in the body, the blood pulsing under your skin alters how light reflects off your tissue.

By tracking wavelength data across 22 facial regions and 8 sections of the palm, the team trained an AI algorithm to calculate pulse waves and flag abnormal blood flow.

In tests run on 215 patients at the University of Tokyo Hospital between 2022 and 2024, the tech performed surprisingly well. The algorithm hit 94% accuracy in identifying stage 1 hypertension (130/80 mm Hg or higher) compared to a continuous blood pressure monitor.

When detecting blood pressure above standard Japanese normal levels, a 30-second recording delivered 86% accuracy, while a 5-second clip still managed 81%.

Even more startling was diabetes screening. Compared against HbA1c blood test results, the camera and algorithm matched diabetes diagnoses with 75% accuracy. Uchida pointed out that since peripheral vascular damage is a hallmark of diabetes, blood flow patterns in the hands and face leave readable digital traces.

External experts are keeping their excitement measured. Dr. Eugene Yang from the University of Washington noted that even the continuous blood pressure cuff used as a baseline benchmark lacks standardized clinical validation protocols for this type of comparison. Furthermore, the dataset was predominantly Asian, filmed in controlled clinic lighting while participants sat completely still.

Why it matters

We think the real promise here isn't replacing clinical diagnostic gear, but practically eliminating the friction of everyday health screening.

Right now, wearable makers have spent nearly a decade trying to crack non-invasive glucose and cuffless blood pressure from smartwatches, running into constant walls with optical wrist sensors. Moving the sensors to a mirror or ambient room camera and letting computer vision do the heavy lifting offers a completely different design route.

Imagine a bathroom mirror equipped with an inexpensive narrow-spectrum sensor that scans you for five seconds while you brush your teeth. It wouldn't need to give you an exact lab number; it just needs to warn you that your microcirculation looks degraded compared to last month.

Still, you shouldn't expect an iPhone app update anytime soon. A 75% accuracy rate for diabetes is far too noisy for medical confirmation, and dealing with varying skin tones, moving heads, and dim indoor lighting will take years to refine. But as a proof of concept, looking at a lens looks a lot better than getting pricked by a needle.