AHA 2024: AI spots high blood pressure, diabetes in slow-motion video

A preliminary study of an artificial intelligence-powered program found that it could detect if a person had high blood pressure from five to 30 seconds of slow-motion video.

The developers from the University of Tokyo said the technology could potentially lead to a non-contact method of screening for hypertension, as well as Type 1 or Type 2 diabetes, as the conditions each subtly alter the flow of blood through the face and the hands.

Researchers said they were able to capture pulse waves under the skin from recordings taken at 150 frames per second, without requiring any wearable devices, sensors or blood tests. The study’s results were presented at the annual scientific sessions of the American Heart Association, being held in Chicago.

“This method may someday allow people to monitor their own health at home and could lead to early detection and treatment of high blood pressure and diabetes in people who avoid medical exams and blood tests,” study author Ryoko Uchida, a project researcher in the department of advanced cardiology at the University of Tokyo, said in a statement.

By videotaping people while they used a continuous blood pressure monitor, the AI algorithm proved to be 94% accurate at spotting stage 1 hypertension, according to the AHA’s guidelines listing measurements of 130 over 80 mmHg.

The study included 215 participants, with an average age of 64. While 62 people had readings of 130/80 mmHg or higher, another 65 came in under that level but above the Japanese standard of 115/75 mmHg.

In addition, compared to blood tests to check glucose levels, the approach was 75% accurate in identifying people with diabetes.

“I was surprised about the applicability of the blood flow algorithm to detect diabetes. However, some of the major complications of diabetes are peripheral neuropathy—weakness, pain and numbness, usually in the hands and feet—and other diseases related to blood vessel damage. It makes sense that changes in blood flow would be a hallmark of diabetes,” said Uchida.

The researchers said future work will require tuning the algorithm to account for potential irregular heartbeats, in addition to developing its compatibility with more affordable cameras, and ultimately validating the technology in a larger study.

“Once it reaches that stage, it may be added to smartphones (or even hung on a mirror where someone sits for a few moments), may be mass-produced and inexpensive,” Uchida said.