Donate to Science & Enterprise

S&E on Mastodon

S&E on LinkedIn

S&E on Flipboard

Please share Science & Enterprise

Phone Camera, A.I. Detect Early Diabetes

Diabetes word cloud

(905513, Pixabay)

18 Aug. 2020. Researchers devised techniques with a smartphone app, and data analyzed with artificial intelligence, to identify people in early stages of type 2 diabetes. A team from University of California in San Francisco describes the techniques and findings in yesterday’s issue of the journal Nature Medicine (paid subscription required).

Diabetes is a chronic disorder where the pancreas does not create enough insulin to process the sugar glucose to flow into the blood stream and cells for energy in the body. In type 2 diabetes, which accounts for at least 90 percent of all diabetes cases, the pancreas produces some but not enough insulin, or the body cannot process insulin. According to the International Diabetes Federation, diabetes affects an estimated 463 million people worldwide, of which 51 million are in North America.

The UC-San Francisco team led by cardiology professor Geoffrey Tison and medical data scientist Kirstin Aschbacher is seeking better tools to detect type 2 diabetes earlier in individuals. The disease, particularly in its early stages, has few noticeable symptoms and can go undetected for years, doing damage to the heart, kidneys, nervous system, and other organs. The authors cite data showing half of the people with diabetes are unaware of their condition and risks to their health.

Tison, Aschbacher, and colleagues hypothesized that data captured by widely available smartphone apps and fitness trackers could provide the raw data needed to screen for undiagnosed type 2 diabetes. Their proposed technique uses images that capture blood flow, in a process known as photoplethysmography or PPG. Plethysmography measures changes in an organ’s volume, such as blood flow, and PPG extends that process with images for capturing the measurements. The researchers then constructed an algorithm to detect and measure diabetes-related indicators in those PPG images.

Colleagues at UC-San Francisco are already investigating use of mobile apps and sensors, with online data collection, in a large-scale study of heart disease, called the Health eHeart study. The team started its inquiry with data from that study, derived from the Instant Heart Rate app offered by medical technology company Azumio, in Redwood City, California. With Instant Heart Rate, users press their finger tip against the phone’s camera, which captures blood flow, and an internal algorithm measures and records heart rate for the individual.

The UC-San Francisco team used nearly 3 million PPG records from some 54,000 individuals to train its deep neural network algorithm. A deep neural network, also known as deep learning, is a form of machine learning where the algorithm gains knowledge from large numbers of cases and becomes more refined as more cases are processed. With deep neural networks, the algorithm is structured in layers, with more layers added for greater sophistication, which enables predictive conclusions, in this case probability of type 2 diabetes.

The researchers validated the algorithm with Instant Heart Rate records from a separate group of 7,800 persons. The team then recruited 181 individuals from three clinics to test the technique. The results show the algorithm alone, using data from the Instant Heart Rate app, could accurately predict diabetes in 81 percent of cases. Adding demographic data and body mass index raises the accurate prediction rate to 83 percent. Scores from the algorithm also strongly correlate with hemoglobin A1c, the standard measure of blood glucose levels.

The authors believe a simple screening tool using mobile technology could greatly improve diabetes detection, particularly among hard-to-reach populations. “The ability to detect a condition like diabetes that has so many severe health consequences using a painless, smartphone-based test raises so many possibilities,” says Tison in a university statement. “The vision would be for a tool like this to assist in identifying people at higher risk of having diabetes, ultimately helping to decrease the prevalence of undiagnosed diabetes.”

More from Science & Enterprise:

*     *     *

Comments are closed.