Wearables can detect early signs of diabetes using machine learning

Artificial intelligence, machine learning, and neural networks are the buzzwords of the tech industry and are often used to make lives more convenient, either for consumers or for the companies selling them products and services. But these technologies are now being utilized to also help make people healthier and even save lives. A recent study made by health startup Cardiogram and the University of California San Francisco revealed that even your common fitness-centric wearable is capable of detecting early signs of diabetes. With a little help from neural networks, of course.

Diabetes is a growing concern in the US, with as much as 100 million believed to be living in prediabetes or already have it. A quarter of those remain undiagnosed throughout their life. Worse, nearly 90% of prediabetics are completely unaware of the danger that lurks behind their seemingly normal lifestyles.

Developing devices to monitor glucose levels is nothing new, but the ideal non-invasive monitor is still a dream. Additionally, most of these devices cater to users who already have diabetes or at least know their condition. It turns out, it's not that hard to know yours of you have an Apple Watch, Android Wear smartwatch, Fitbit, or any wearable that has a heart rate monitor.

This seemingly magical way of detecting early signs of diabetes works based on how heart rate variability can be connected to the likelihood of diabetes. We have the formula but not the data to feed into it. That's where machine learning comes in. Since it's logistically impossible to gather the millions of data needed to train a neural network, Cardiogram and UCSF resorted to semi-supervised machine learning techniques to train the DeepHeart deep neural network using 33,628 person-weeks worth of health sensor data. They later verified DeepHeart's accuracy against a separate data set of 12,790 person-weeks and got an 85% success rate.

The best part about this study is that anyone with a compatible smartwatch or fitness tracker can reap the benefits of this neural network. All they need is to download the Cardiogram app from the Apple App Store or Google Play Store. The app works on all Apple Watch models and Android Wear smartwatches with hear rate sensors. In addition to early signs of diabetes, DeepHeart has also been demonstrated to detect other heart-related conditions like hypertension, sleep apnea, atrial fibrillation.