The scourge of Type 2 diabetes (T2D) continues to blind, maim, and cause premature death for a growing number of people worldwide. From 2000 to 2016, diabetes caused a 5% increase in premature mortality, according to the World Health Organization. During that period, cases of diabetes increased faster in developing countries than industrial nations due to westernized diets and decreasing physical activity. Early detection of impending diabetes can help delay or prevent the onset of the disease, which fuels research worldwide in attempts to develop new detection technology.
We have written about multiple institutions and groups that are developing technologies that combine biomarker sensing with artificial intelligence to detect early signs of diabetes or help with treatment for the disease. For example, we wrote about a group at Harvard that showed machine learning could predict which T2D patients will have the greatest problems with glycemic control. SweetchHealth and WellSpan Health demonstrated an AI platform that can detect diabetes using data from smartphones and wearables.
Researchers at the University of California San Francisco recently published a study in Nature that showed a smartphone camera successfully identifies a digital biomarker of diabetes. The UCSF team taught a deep neural network (DNN) to detect diabetes using photoplethysmography (PPG). The group used an initial training group of nearly 54,000 individuals and validated the results with a second group consisting of 7,800 people and a third group of 181 patients from three clinics. With both groups, the DNN accurately interpreted the biomarkers in approximately 75% of the cases with a high degree (95%) of confidence. Based on a significant positive association between the DNN scores and hemoglobin A1c scores, the UCSF team stated that smartphone-based PPG can function as an easily attainable, non-invasive diabetes detection tool.
Note that this system does not measure blood glucose levels; it simply is a screening tool that determines whether the patient has diabetes. It’s safe to assume at this juncture that continued work with artificial intelligence, machine learning, and deep neural networks to detect early signs of diabetes will result in faster identification and treatment. It’s a shame to direct these resources at a largely preventable disease, but the potential for relief outweighs any wishes that people, including myself, had better nutritional habits.
