Smartphone-Measured Photoplethysmography Serves as Digital Biomarker of Diabetes

A noninvasive digital biomarker is being developed at the University of California, San Francisco (UCSF) for detecting Type 2 diabetes using a smartphone camera and deep learning algorithm. This innovation could provide a low-cost, in-home alternative to blood draws and clinic-based screening tools.

The researchers hypothesized that a smartphone camera could be used to detect vascular damage due to diabetes by measuring photoplethysmography (PPG) signals, which most mobile devices, including smartwatches and fitness trackers, are capable of acquiring. They developed a deep neural network (DNN) to detect prevalent diabetes using smartphone-based PPG signals. The smartphone flashlight and camera measured PPGs by capturing color changes in the fingertip corresponding to each heartbeat.

The DNN takes a PPG waveform as the sole input and provides a score between 0 and 1, with higher scores suggesting greater likelihood of prevalent diabetes. The DNN accepts PPG waveforms ~21 seconds in duration, but shorter or longer durations can be padded or cropped, respectively. PPG waveforms were obtained by placing the index fingertip on the smartphone camera using the Azumio Instant Heart Rate smartphone application (Azumio Inc). Changes in reflected light intensity recorded by the smartphone camera were interpreted as pulsatile blood volume changed. PPG waveforms were collected at either 100 or 120 Hz. Courtesy of University of California, San Francisco.

The researchers obtained nearly 3 million PPG recordings from 53,870 patients who used the Azumio Instant Heart Rate app on an iPhone and reported having been diagnosed with diabetes by a health care provider. This data was used to both develop and validate the deep-learning algorithm used to detect the presence of diabetes through smartphone-measured PPG signals.

Overall, the algorithm correctly identified the presence of diabetes in up to 81% of patients in two separate data sets. In an additional data set of patients enrolled from in-person clinics, it correctly identified 82% of patients with diabetes.

Among the patients that the algorithm predicted did not have diabetes, 92% to 97% indeed did not have the disease across the validation data sets. When this PPG-derived prediction was combined with other patient information such as age, gender, body mass index, and race/ethnicity, predictive performance improved further.

Smartphone-based photoplethysmography could provide a noninvasive digital biomarker of prevalent diabetes. Such a screening tool could be deployed easily, using technology already contained in smartphones, and could extend screening to populations out of reach of traditional medical care.

“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,” Dr. Geoffrey H. Tison said. “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.”

The researchers recommend further study to determine the effectiveness of this approach for specific clinical applications, such as screening or therapeutic monitoring.

The research was published in Nature Medicine (www.doi.org/10.1038/s41591-020-1010-5).