Algorithm uses smartphone camera, flashlight to spot diabetes

The tool considers vascular changes and individual risk data to make its predictions.
By Dave Muoio
Share

New research set to be presented next week at the describes how the camera on a smartphone could be used to screen for Type 2 diabetes.

Diabetes is associated with changes to how blood flows through a person’s blood vessels, an effect that researchers from UCSF suggest could be used to detect early changes. By applying a deep learning algorithm to an existing heart rate dataset, the team’s approach was able correctly identify the condition in nearly three-quarters of individuals examined in the study.

“[Photoplethysmography (PPG) signal is] a measurement that's already readily obtained from smartphones and wearable devices to track heart rate," Dr. Robert Avram, post-doctoral fellow at UCSF Medical Center and the study's lead author, said in a statement. "We've demonstrated that by using deep learning and a smartphone camera alone, we can also detect vascular changes associated with diabetes and with reasonable discrimination.”

According to a release detailing the research, Avram and colleagues collected data from 54,269 adult individuals who had enrolled in . These participants had measured their heart rates with the Azumio Instant Heart Rate smartphone app, which uses the phone’s camera and flashlight to measure blood volume variations in the user’s fingertips. The study’s participants were aged an average of 45 years, and 7 percent self-reported a diabetes diagnosis.

After applying the algorithm to PPG signal data alone, the researchers’ model correctly identified 72 percent of the participants with diabetes and increased this rate to 81 percent when incorporating other collected risk factors. Additionally, the model correctly predicted the absence of diabetes for 97 percent of those without the condition.

Why it matters

Since the risk factor-augment algorithm’s success rate is roughly on par with diabetes risk scores currently being employed by providers, Avram and colleagues are currently applying their model in two cardiovascular prevention clinics. Validating the digital tool would allow patients to easily gauge their own diabetes risk and, as a result, more promptly seek appropriate care.

"Diabetes can be asymptomatic for a long period of time, yet adverse vascular changes still occur silently, which can lead to cardiovascular complications. This makes it especially important for us to examine low-cost, noninvasive opportunities that make it easy to screen millions of people. To date, a noninvasive, widely-scalable screening tool for diabetes has been lacking," Avram said. "Based on our findings, this strategy could become a low-cost way to screen for diabetes at home because it can be derived from any optical system that has a camera and a flashlight, and most people have a smartphone.”

What’s the trend

Newer diabetes technologies designed for consumer use are generally geared toward managing the chronic condition, but the diagnostic and risk-estimate space has lately seen some advancements as well. Artificial intelligence screens for diabetic retinopathy in particular have gained some ground, with both Idx and Google’s tools earning themselves plenty of headlines. And just this weekend, 23andMe launched a genetic predisposition test for Type 2 diabetes that is already rousing some debate from the company’s critics.