One of our favorite themes here at Health Tech Insider is how the “quantified self” movement is going to produce mountains of Big Data that can be mined to find useful correlations. This will lead to earlier diagnosis and more effective treatment of disease, which will save lives, improve the quality of life, and create huge reductions in healthcare costs. But first you need the data.
One problem is that much of the data is not very good. There can be a lot of variation in the raw data, which engineers often refer to as “noise.” This noise can obscure the meaningful information contained within the numbers. Instead of black and white results, you get a jumbled gray mess. Fortunately, mathematical algorithms can do wonders at extracting the needle of useful information from a haystack of data.
Cardiogram is a company that has developed a smartphone app (iOS and Android) designed to collect data from wearable devices with heartbeat sensors. They use deep machine learning to identify cardiac information in the noisy data from wearable devices. In partnership with University of California San Francisco Cardiology, they used clinical data to train a neural network system to extract the heart rate data. Early tests with the UCSF Cardiology’s Health eHeart Study indicate that the system may ultimately be able to reach 90% accuracy, even using “consumer grade” wearables.
In the end, there are two ways to get better data. The first is to create more accurate sensors, but the other is to use more intelligence to interpret the data that is gathered. It may well be that the second path will be less expensive, easier to implement, and in the end, produce as good or better results.
