AI-Driven Blood Pressure and Diabetes Screening Shows Promise in New Study
AI-powered system could enable non-invasive screenings for blood pressure and diabetes via high-speed video capture, reducing the need for blood tests.
A new AI-driven diagnostic system may soon enable non-contact health screenings for high blood pressure and diabetes, based solely on high-speed video. According to research led by Ryoko Uchida, B.Sc. (Pharm.), from the University of Tokyo’s Department of Advanced Cardiology, this technology uses an algorithm capable of detecting high blood pressure and diabetes by analysing subtle blood flow changes in the skin. The preliminary study, which will be presented at the American Heart Association’s Scientific Sessions 2024, is already attracting attention for its potential to simplify and democratise health monitoring.
“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,” Uchida explained.
The AI-based system captures rapid video sequences of a person’s face and hands, measuring blood flow patterns that correspond with specific health markers. The system demonstrated an impressive 94% accuracy in detecting hypertension and a promising 75% accuracy rate in identifying diabetes when cross-checked against traditional clinical tests.
High-Accuracy Detection with Short Screening Times
The study involved 215 adults, mostly Japanese and other Asian individuals, who underwent video-based analysis and traditional tests for hypertension and diabetes. Each participant’s face and palms were recorded at 150 frames per second, and the AI algorithm then examined the video footage to discern blood flow characteristics associated with hypertension and diabetes.
Results show that this video-based system closely aligns with established clinical measures. For hypertension, the AI reached a 94% accuracy when identifying blood pressure levels above 130/80 mm Hg, using the American Heart Association’s guidelines as a benchmark. Even with a shorter screening time, the system maintained high levels of precision: a 30-second video clip achieved 86% accuracy, while a 5-second clip retained 81% accuracy in detecting elevated blood pressure.
Diabetes detection, though slightly less precise, still produced a promising outcome with 75% accuracy when compared to HbA1c blood tests, which reflect average blood glucose levels over one to two months. Uchida noted that the applicability of the blood flow algorithm for diabetes screening was unexpected, attributing the algorithm’s success to its ability to detect blood vessel changes typically linked with diabetes complications, such as peripheral neuropathy and blood vessel damage. “It makes sense that changes in blood flow would be a hallmark of diabetes,” she added.
Pathway to Clinical and Commercial Viability
While the current study demonstrates significant potential, Uchida acknowledges that further refinement and validation of the technology are necessary. Key areas for improvement include incorporating an algorithm that accounts for arrhythmias or irregular heartbeats, which could affect blood pressure readings. “In the future, the prototype camera we used to develop the algorithm could be substituted with an affordable sensor that uses only the essential wavelengths and requires just a few seconds to gather data,” Uchida explained. She envisions that the system could eventually be integrated into smartphones or household mirrors, making health screening as accessible as taking a selfie.
Despite the promising results, some experts remain cautious about the wide-scale application of this technology. Eugene Yang, M.D., M.S., from the University of Washington, notes that while the AI algorithm and video-based screenings show potential, the lack of validation protocols for these devices is a critical issue. Yang advises that until such protocols are standardised, traditional devices for measuring blood pressure and glucose should remain the primary diagnostic tools. “Until we have approved validation protocols for these technologies, including wearable devices like smartwatches, we must use validated devices for measuring blood pressure and glucose levels,” he said.
The study also has limitations, with findings derived predominantly from Japanese and other Asian participants, raising questions about its applicability across diverse populations. Moreover, the technology was tested in controlled hospital environments, which might differ from real-world settings with varying lighting and movement levels.
Future Potential and Regulatory Considerations
Looking ahead, Uchida’s team intends to refine the system’s accuracy, especially for diabetes detection, before seeking regulatory approval. The ultimate goal is to create an accessible, non-invasive diagnostic tool approved by regulatory bodies, potentially enabling widespread use in everyday settings. “Currently, the only way to confirm the diagnosis of diabetes is invasive blood tests,” Uchida stated, “however, if it were to require only a non-invasive photo or video, that could be a game-changer.”
The research reflects a growing trend towards non-invasive health diagnostics, especially for high-risk conditions like hypertension and diabetes, which are major risk factors for cardiovascular disease. The application of this AI-powered video technology could mark a significant shift towards user-friendly, at-home health monitoring tools. However, as Yang highlights, the journey from promising research to validated clinical application is complex and demands rigorous testing.
