A short video of a person’s face may one day help doctors identify two of the world’s most common chronic diseases.
Researchers in Japan have developed an artificial intelligence system that can look at tiny changes in facial blood flow and use them to screen for high blood pressure and diabetes.
The technology does not require a blood pressure cuff, a blood sample or a wearable device. In the study, recordings lasting only five to 30 seconds produced promising results, suggesting that contactless health checks could eventually become possible in clinics, pharmacies or even other everyday settings.
The research was carried out by scientists from the University of Tokyo and the Institute of Science Tokyo in Japan. The latest findings are being presented at ESC Congress 2026, the annual scientific meeting of the European Society of Cardiology.
High blood pressure, also called hypertension, is extremely common but often causes no obvious symptoms. Around 1.4 billion adults between ages 30 and 79 are estimated to have hypertension worldwide, and many do not know they have it.
Diabetes is another major global health problem, affecting about 589 million people. When diabetes is not diagnosed and controlled, high blood sugar can gradually damage blood vessels, nerves, kidneys, eyes and other organs.
Both conditions also increase the risk of cardiovascular disease. Finding them early gives people a chance to receive treatment and make lifestyle changes before serious complications such as heart attack, stroke, kidney disease or vision problems develop.
However, screening large numbers of people can be difficult. Blood pressure normally needs to be measured with a cuff, while diagnosing diabetes usually involves blood tests, and people who rarely visit a doctor may therefore remain undiagnosed for years.
The Japanese researchers wanted to know whether information already visible in the skin could provide another way to identify people who may need further testing. Their idea was to combine a special camera with machine learning, a form of AI that learns patterns from large amounts of data.
The prospective study included 215 people, consisting of patients who had already been diagnosed with disease as well as healthy volunteers. Each participant had a brief, high-speed recording taken of the face and palms with a spectroscopic camera.
Unlike an ordinary camera, a spectroscopic camera can capture information about light at different wavelengths. These signals can reveal very small changes linked to blood moving through vessels beneath the skin, even when those changes are difficult for the human eye to see.
The AI system examined several types of information in the videos. These included pulse-wave patterns related to how blood travels through arteries, changes in skin blood flow and characteristics of skin color measured across different parts of the light spectrum.
The participants also received conventional medical assessments for hypertension and diabetes. This allowed the researchers to compare the AI system’s results with established clinical measurements.
Earlier work from the research group showed that a 30-second recording of the face and palms could identify hypertension with 95.0% accuracy. The system correctly recognized normal blood pressure in 100.0% of cases and detected hypertension with a sensitivity of 89.2%.
The researchers also tested whether the recording could be made much shorter. With only five seconds of video, the hypertension screening method still achieved an accuracy of 90.3%, an encouraging result for a technology designed to be quick and easy to use.
The new analysis focused on diabetes. By examining blood flow patterns in facial video, the AI system detected diabetes with 88.2% accuracy when it used a 30-second recording and 81.2% accuracy when the recording lasted only five seconds.
The researchers went a step further and tested whether facial video alone could estimate actual blood pressure without a cuff. For systolic blood pressure, which is the top number in a blood pressure reading, the average percentage error was 8.6%.
The average difference between the AI estimate and the conventional measurement was minus 2.6 millimeters of mercury, or mmHg. That average error fell within one widely used technical standard, but the amount of variation between individual measurements was still too large to meet the full standard.
This limitation is important because a screening tool can look impressive when results are averaged across a group while still being inaccurate for some individuals. The researchers therefore say the technology needs to be tested in larger studies involving people from different backgrounds and multiple medical centers.
The results are nevertheless notable because the system found useful health signals without touching the participant. If future studies confirm the findings, a camera-based check could potentially flag people who should have a proper blood pressure measurement or diabetes test.
Such a system would be especially useful as a first screening step rather than a replacement for medical diagnosis. A person identified as being at higher risk would still need standard clinical testing before being diagnosed or treated.
The approach may also help reach people who do not regularly attend health checks. Fast screening in convenient locations could encourage earlier follow-up and may reduce the number of people living unknowingly with hypertension or diabetes.
There are also practical questions that remain unanswered. The study involved only 215 participants at a single center, so researchers still need to know how well the system performs across different ages, skin tones, lighting conditions, camera types and health conditions.
Overall, the study provides an interesting proof of concept rather than a ready-to-use medical test. Accuracy above 80% to 90% from very short videos is promising, but the remaining errors matter when a tool is being used to identify disease.
The strongest potential use may be broad, low-effort screening that directs people toward proper medical assessment. If larger and more diverse studies reproduce the results and the blood pressure estimates become more consistent, facial-video AI could become a useful addition to traditional screening rather than a substitute for it.
If you care about high blood pressure, please read studies about vitamins impacts on high blood pressure people need to know, and how to manage high blood pressure and diabetes with healthy foods.
For more health information, please see recent studies about the best and worst foods for high blood pressure, and modified traditional Chinese cuisine can lower blood pressure.
Source: University of Tokyo and Institute of Science Tokyo.
