Marriage of AI, Google chips will save diabetics from a lot of pricks
Blood-scanning radar could be built into a smartwatch
AI may help people with diabetes monitor their blood sugar levels without puncturing their own skin, according to a research from the University of Waterloo.
Most diabetics have to prick their fingers multiple times a day to draw a drop of blood to be dabbed onto a strip of test paper, before it is fed into a blood glucose meter. But the new system could eliminate that invasive process.
"We want to sense blood inside the body without actually having to sample any fluid," said George Shaker, an engineering professor who leads a the project. "Our hope is this can be realized as a smartwatch to monitor glucose continuously."
The idea is based around a sensor Google and Infineon, a German hardware company, developed. In 2016, Google showcased Project Soli, a chip smaller than a quarter, that emits high-frequency radio waves at 60-GHz millimeter-wave band and measures reflected signals bouncing off a user's hand to track its motion. It’s designed to fit into wearables, phones, computers, cars and IoT gizmos.
Researchers believe that this can be applied to sense the motion of blood by beaming radar signals to a person’s finger, according to a paper published in the International Journal of Mobile Human Computer Interaction. They won a research grant under Google’s ATAP Alpha program, giving them access to the device and its developer kit.
Ayyy-EYE! Google code 'predicts heart disease' by eyeballing retinasREAD MORE
Previous experiments showed that the level of electrical resistance in blood changed depending on the amount of sugar it contained. Glucose changes the absorption and reflection properties of blood and that can be measured in the reflected wave.
That data is then processed as multidimensional matrices and fed into a random forest classifier algorithm. “[It] inspects more than 500 features, such as the reflected signal amplitude, phase, delay, energy, and dispersion,” George Shaker, lead-author of the paper and an adjunct assistant professor at the University of Waterloo, told The Register.
After successful initial experiments, the team moved onto using fake blood solutions with different glucose concentrations to teach the system what features are correlated with certain sugar levels.
After training, the team tested their system on ten volunteers and it was around 85 per cent accurate. The tests were performed on liquid blood samples, however, and the system is not yet ready to measure glucose levels through the skin. "We have shown it is possible to use radar to look into the blood to detect changes," Shaker said.
The next step is to work out how to process the information on the chip instead of relaying it to a computer to get results. The researchers are also collaborating with Infineon to try and shrink the device to fit into a smartwatch.
"I'm hoping we'll see a wearable device on the market within the next five years," said Shaker. "There are challenges, but the research has been going at a really good rate." ®