No-contact AI Tool Shows Potential for Detecting Diabetes and High Blood Pressure
Key takeaways:
- Researchers in Japan developed a system that can diagnose diabetes and hypertension using video and artificial intelligence.
- Small changes in blood flow can be seen in the face and palms of people with these conditions.
- More research is needed to determine if the results would be accurate in the general population.
A process that uses artificial intelligence to analyze high-speed video can accurately diagnose diabetes and high blood pressure, according to research conducted in Japan.
The findings were recently presented at the American Heart Association's scientific conference, though have yet to be published in a peer-reviewed medical journal.
Developed at the University of Tokyo, the system doesn't require blood testing or a blood pressure cuff. Instead, a high-speed video records the hands and face of the person being screened. The camera records at a rate of 150 images per second, about five times faster than a typical video recording. The video images are then reviewed by artificial intelligence using an algorithm that detects blood flow features in the skin.
Blood flow to the hands and face is affected by diabetes and blood pressure in subtle ways that can be detected in video of a person's face and palm, said study author Ryoko Uchida, a researcher in the department of advanced cardiology at the University of Tokyo.
The system accurately identified type 1 and type 2 diabetes 75% of the time compared with an A1C test. When used to detect stage 1 hypertension (130/80 mm Hg or higher), the approach was 94% accurate compared to a monitor that recorded blood pressure at the same time the video was taken.
Dr. Eugene Yang, professor of medicine at the Washington School of Medicine in Seattle, who was not involved in the study, said ways to non-invasively detect diabetes and high blood pressure were a welcome development in reducing risk factors for heart disease.
It’s important to note that this study was conducted using 200 participants who identified as Asian, meaning it's unclear how the results might translate to a larger number of participants and non-Asian populations using the same diagnostic process.
"The technologies rely on signals generated from the surface of the skin," Yang said. "Differences in skin tone, especially darker skin tones, may make it more difficult to extract information to estimate blood pressure or glucose levels. The study needs to include adequate numbers of people of all races and ethnicities to ensure that the algorithms perform equally well."
Yang also noted the study devices used to detect high blood pressure were FDA-approved, but needed to be validated for accuracy.
The bottom line
In a small study, an AI-based process has demonstrated the ability to diagnose diabetes with 75% accuracy and hypertension with 94% accuracy by noting changes to blood flow in the face and palms. Video samples were taken in cardiology wards and outpatient booths at the University of Tokyo Hospital. More research is needed to evaluate the process used in this study in larger and more diverse populations.
It is possible that this app in the future could benefit people who typically avoid medical exams and blood tests. The researchers plan to continue to refine the algorithm's ability to detect diabetes, with the hopes of seeking FDA approval in the future.
Learn more about diabetes screening and prevention here: