Key takeaways

• This study explored the capacity of a machine-learning algorithm to identify hypertension and diabetes using facial videos.
• The algorithm achieved over 90% accuracy for detecting hypertension and over 80% for diabetes from just 5-second videos.
• If proven effective, this AI model could serve as a scalable option for broader screening.

Munich, Germany – 26 August 2026: A study to be presented at ESC Congress 2026 suggests that AI-driven analysis of facial video images can efficiently and accurately identify undiagnosed high blood pressure and diabetes.

There are approximately 1.4 billion adults worldwide between the ages of 30 to 79 with hypertension and about 589 million people living with diabetes. Both conditions are significant, modifiable risk factors for cardiovascular disease, yet many individuals remain undiagnosed. Current screening methods typically require visits to clinics or the use of wearable devices, which limits their reach.

Researchers from the University of Tokyo and the Institute of Science in Tokyo examined whether the analysis of facial videos through AI could enhance the diagnosis of hypertension and diabetes. The lead presenter, Ms. Ryoko Uchida, stated, “Our goal was to develop an AI algorithm to facilitate contactless screening in everyday scenarios, enabling earlier and broader detection of common health issues.”

The study involved 215 participants, including diagnosed patients and healthy individuals, all recorded for a brief 30-second high-speed video session of their face and palms with a spectroscopic camera. A machine-learning algorithm processed these videos, analyzing data such as pulse-wave dynamics, skin blood flow patterns, and skin color spectral features. Participants also underwent conventional assessments to pinpoint hypertension and diabetes.

Previously published results indicated that the algorithm could identify hypertension with a striking 95% accuracy from a 30-second recording, with a 100% sensitivity for normal blood pressure and an 89.2% sensitivity for hypertension. The accuracy dropped slightly to 90.3% using a 5-second video.

In new findings, based on patterns of facial blood flow, the algorithm maintained a high accuracy for diabetes detection, achieving 88.2% from a 30-second video and 81.2% from a 5-second video.

Remarkably, the algorithm also estimated blood pressure from a facial video alone, without any cuff. The average absolute percentage error for systolic blood pressure (SBP) was 8.6%. While the mean error for SBP was −2.6 mmHg—a figure that fits within the AAMI’s guideline of ±5 mmHg—the standard deviation error was ±12.0 mmHg, surpassing the AAMI criterion of ±8.0 mmHg. Future research is intended to improve this variability using larger, more diverse datasets and enhanced feature optimization.

Ms. Uchida concluded, “Our machine-learning algorithm accurately identified hypertension and diabetes from facial spectroscopic video records as brief as 5 seconds. We aspire to confirm these results with larger cohorts in varied populations to enable real-world applications. If validated, this approach could facilitate screenings in everyday environments—without the need for cuffs, blood samples, or specific clinic visits—helping to identify individuals at risk who might otherwise go undetected and untreated.”

“It’s quite impressive that AI-augmented technologies are developing such effective tools for early disease prevention,” commented Associate Professor Nico Bruining, Programme Co-Chair of the ESC Digital and AI Summit and Editor-in-Chief of the European Heart Journal – Digital Health. “This method’s speed, ease, and contactless nature means it could be applied in various settings beyond hospitals, potentially reaching far more individuals than current screening techniques. Early detection allows for timely treatment and lifestyle adjustments, which can prevent heart attacks, strokes, and other cardiovascular issues.”

The latest developments in AI and cardiovascular healthcare will be showcased at the ESC Digital & AI Summit in Basel, Switzerland, on 12–13 November 2026. The event promises an in-depth look at many innovative AI solutions reshaping cardiovascular prevention and care. It will gather a growing community of clinicians, researchers, innovators, and industry leaders dedicated to advancing cardiovascular health.

ENDS