FDA authorizes first AI-powered armband for COVID-19 screening

The FDA has authorized the use of its first digital, machine-learning-powered device to help screen people for COVID-19.

Tiger Tech Solutions’ wearable monitor is strapped to the upper arm and uses light sensors to sense blood flow, similarly to many consumer electronics and fitness trackers. Within three to five minutes, it uses an artificial intelligence model to crunch the data on the person’s pulse rate and other factors to determine whether their blood could be clotting more easily than normal.

This state of hypercoagulation, among other signs, has been linked to coronavirus infections—and, when combined with temperature checks, it could help spot people over the age of 5 who are carrying the virus without showing any symptoms. 

The armband device is not designed to take the place of a diagnostic test but to provide a second-line screening option for hospitals and elsewhere if a person is not displaying a high fever.

The COVID Plus armband
monitor (Tiger Tech)

“Combining use of this new screening device, that can indicate the presence of certain biomarkers, with temperature checks could help identify individuals who may be infected with the virus, thus helping to reduce the spread of COVID-19 in a wide variety of public settings, including healthcare facilities, schools, workplaces, theme parks, stadiums and airports,” said the FDA’s device center director, Jeff Shuren, M.D., in an agency statement.

Machine learning has been explored as a way to screen for COVID-19 for the past year—especially to help read chest X-ray and CT scans for signs of the virus in the lungs—however, a recent evaluation by researchers at the University of Cambridge found most of these models are not yet up to snuff.

They discovered that out of more than 300 computer models described in scientific papers, none of them were suitable for detecting infections from digital medical images, largely due to methodological flaws and a lack of reproducibility as well as biases in poorly assembled or too-small data sets. Their findings were published in Nature Machine Intelligence.

For example, this includes studies that trained the algorithm on infection-free images taken from scans of children and compared them to positive scans from adults—resulting in AI that could mainly tell the difference between children and adults, instead of who has the disease, as children are far less likely to have COVID-19.

According to the FDA, Tiger Tech’s COVID Plus Monitor matched up with the findings of diagnostic tests 98.6% of the time in finding positive cases in a hospital and 94.5% of the time when ruling out negative cases. It showed similar performance when used in the school setting.