Information for Business from Lenovo
Contributor: Orin Thomas
Unlocking the potential of advanced health analytics

Devices with sensors that can monitor health telemetry are allowing us to gain more insights about our health.

One of the big booms in the past few years has been in devices you strap to your body, such as Fitbit, Apple Watch and Microsoft band. These include sensors that can monitor health telemetry, and each year, these devices have become more sophisticated.

Initial versions would measure telemetry such as the number of steps the wearer had taken and the wearer’s heart rate. Current versions measure telemetry including skin temperature, galvanic skin response as well as ultraviolet rays, light, and air pressure. As technology improves, sensors that can check blood sugar levels and blood pressure are also likely to be included in consumer health monitoring devices.

These devices often pair to a smartphone, and through the smartphone, the telemetry data is forwarded to servers on the internet. Short-term data might be stored on the phone, while all data generated by the device can be stored using the greater storage capacity of cloud services on the internet. While the smartphone is able to provide us with beautiful and handy displays about our own health telemetry, it will get increasingly interesting over the next few years when all this telemetry is married with deep learning.

Diving into deep learning

Deep learning allows for the discovery of unknown relationships by analysing immense sets of data. The idea is that if you collect enough reliable data and you throw enough processing power at it, you can discover relationships between different aspects of that data that wouldn’t have been apparent or obvious with much smaller data sets.

In the past, you had to have a hypothesis about a potential relationship and then you had to go out and collect experimental data to determine whether your hypothesis was true or not. Science has always had something of a chicken-and-egg problem when it comes to experiments. You need a hypothesis around which to design an experiment but you don’t get the hypothesis out of thin air. You get a hypothesis because you’ve observed a phenomenon. But how did you know what phenomenon to observe unless you had some form of hypothesis? Most of the time, you can only look for something if you have an idea that you know what you are looking for.

Deep learning sidesteps this philosophical conundrum because rather than focusing in on specific hypothesis suggested data, as is the case with traditional scientific research, it examines massive data sets where the data hasn’t been collected on the basis of a specific hypothesis to determine if relationships exist in the data. With enough data and processing power, it’s possible to find relationships that might never have occurred to even the most diligent of observers.

Microsoft recently found this type of relationship when it came to search terms for its Bing search engine. Microsoft used deep learning techniques to analyse large volumes of search data. They were able to identify people who had been diagnosed with pancreatic cancer on the basis of search terms. They were then able to look back across many months of that person’s search history and find that prior to their formal diagnosis by a doctor, people were querying the search engine for specific symptoms. There was a correlation between people who searched for specific terms using the Bing search engine and a diagnosis, months or years later of pancreatic cancer. You can find out more about this study here.

Health insights from data

If you think about all the health data telemetry that's being collected by devices such as Fitbit, Apple Watch, and Microsoft band, it seems reasonable to assume that deep learning will likely to bring us insights about our health. For example, as these devices become more popular, doctors are using the data they have generated and stored during a check-up. It’s more useful for a doctor to have an understanding of your health telemetry over a period of time than it is checking your blood pressure, pulse and other vital statistics only during the consultation.

It’s likely that through constant monitoring and analysis of health telemetry, health problems may be visible in data before we perceive them ourselves. For example, the heart rate monitor in a device attached to our wrists is far better at detecting abnormalities in our heartbeat than we will be able to without such a device. It’s well known that a person’s chances of surviving a cardiac arrest are related to how quickly they get medical attention. It’s possible at some stage in the future that wearers of health telemetry monitoring devices will get notifications from their phone that they should visit a doctor or an emergency department because their heart-rate telemetry indicates that they are in the preliminary stages of a cardiac arrest, something they would notice physically much later once the symptoms were more pronounced.

Portable monitoring devices, currently only used by technophiles, are likely to become more commonly used over the next decade. As these devices become more sophisticated, they will collect an increasing variety of health telemetry. The more health telemetry they collect, the more likely it is that deep learning analytic techniques will discover more from collected health telemetry, allowing us to know more about our health.

SHARE
Recommended articles
How healthcare will become predictive, personalised and preventative with tech
Matt Meakins