Contactless AI Tool Detects Hypertension and Diabetes from Facial Video

Hypertension and diabetes are widespread yet often undiagnosed, driving preventable cardiovascular morbidity and mortality. Screening typically depends on clinic visits or wearable devices, which can limit reach and delay detection. Health systems need rapid, scalable tools that fit routine patient pathways without contact or specialized equipment. To meet this need, researchers in Japan have developed an AI method that analyzes brief facial videos to identify both conditions and estimate blood pressure.

The machine-learning algorithm was developed by the University of Tokyo and the Institute of Science Tokyo and will be presented at ESC Congress 2026. It processes spectroscopic facial video to detect hemodynamic and microvascular signatures linked to hypertension and diabetes. It is designed for contactless screening that can be completed in seconds.

In a prospective, single-center study, investigators enrolled 215 participants, including diagnosed patients and healthy volunteers. Each subject underwent a short, high‑speed recording of the face and palms using a spectroscopic camera. The algorithm extracted features from pulse‑wave dynamics, skin blood‑flow patterns, and the spectral characteristics of skin coloring, and participants received conventional assessments to identify hypertension and diabetes.

For hypertension detection, the model achieved 95% accuracy from a 30‑second recording that analyzed facial and palm video. Sensitivity for normal blood pressure was 100.0%, while sensitivity for hypertension was 89.2%. Accuracy remained high at 90.3% using only a 5‑second video. For diabetes detection based on facial blood‑flow patterns, accuracy reached 88.2% with a 30‑second recording and 81.2% with a 5‑second recording.

The system also estimated systolic blood pressure from facial video alone with a mean absolute percentage error of 8.6%. Its mean error was −2.6 mmHg, within the Association for the Advancement of Medical Instrumentation (AAMI) limit of ±5.0 mmHg, although the standard‑deviation error was ±12.0 mmHg, exceeding the AAMI criterion of ±8.0 mmHg. Future work will target variability reduction through larger, multicenter datasets and feature optimization.

“Our machine-learning algorithm accurately detected hypertension and diabetes from facial spectroscopic video recordings as short as 5 seconds. We intend to validate these findings in larger cohorts across more diverse populations to support real-world application. If validated, this contactless approach could allow people to be screened in everyday settings—without cuffs, blood sampling or a dedicated clinic visit—helping to identify at-risk individuals who would otherwise remain undiagnosed and therefore untreated,” said Ryoko Uchida, presenter.

“It is remarkable that AI-supported technologies are enabling the development of such powerful tools for early disease prevention. Because this approach is quick, easy and contactless, it could be used in many settings beyond hospitals, giving it the potential to reach far more people than traditional screening methods. Detecting these conditions early means treatment and lifestyle changes can start sooner, helping to prevent heart attacks, strokes and other cardiovascular diseases,” said Associate Professor Nico Bruining, program co-chair of the ESC Digital and AI Summit and editor-in-chief of the European Heart Journal—Digital Health.

Related Links
Institute of Science Tokyo 
University of Tokyo
European Society of Cardiology