It has been rumored more than once that Apple is working on a non-invasive blood glucose meter for Apple Watch. Now scientists have shown that the heart rate sensor in the current generation of watches is able to successfully diagnose diabetes in the early stages.
In one study using Apple Watch and a watch on Android Wear, app developers at Cardiogram and the University of California at San Francisco trained a neural network called DeepHeart to differentiate diabetics from healthy people with precision. 85%.
The study involved 14,011 Cardiogram users. The information obtained thanks to them helped in the training of DeepHeart, which analyzed and compared data from sick and healthy people. Moreover, it was not only about diabetes, but also about hypertension, sleep apnea, atrial fibrillation and high cholesterol levels.
Typical deep learning algorithms require a huge amount of information, millions of labeled examples. However, in medicine, each such example means that a person's life is in danger – for example, these are people who have recently experienced a heart attack. To solve this problem, the researchers used two semi-automatic deep learning techniques, which allowed them to find the use of both tagged and unmarked information to increase accuracy.
This is made possible by the link between diabetes and the autonomic nervous system. As a result, DeepHeart can diagnose diabetes through the readings of the heart rate sensor. In particular, even at an early stage of the disease, the pattern of heart rate variability changes enough for this change to be detected.
As for the non-invasive blood glucose meter for Apple Watch, it will take several years before this technology is implemented. Cardiogram co-founder Brandon Ballinger noted that the company is ready to integrate it into the DeadHeart watch if such a sensor is actually added.
Cardiogram will continue research in this direction in 2018. One of the most important planned changes is the addition of DeepHeart to the application to provide more comprehensive statistics.