How machine learning helps health
There are numerous ways that deep learning, AI and machine learning can boost health, from diagnosis to treatment. Here are five projects already in the works that could change how we identify and treat illness.
Ada
Ada is a smartphone app that “talks to you like a really good doctor would,” taking into account your history and symptoms. It’s now being ported to Amazon’s Echo, so you can ask such questions to the Alexa voice assistant, too. Rather than act like a symptom checker online, each aspect of a question is actually computed for results, offering advice and next steps – including whether you should go and see your human doctor. Naturally, there are disclaimers: Ada is not a medical diagnosis, so see a doctor if you’re worried.
DeepMind and Moorfields
DeepMind’s kidney app was its first project; the second is a partnership with Moorfields Eye Hospital, which aims to help optometrists detect common eye disorders by letting deep-learning systems look at scans. It takes a long time for humans to read the complex eye scans, but DeepMind’s system will train using a million images, with the aim of reducing the time doctors spend staring at images, instead boosting their time with patients.
OxSight
Spun out of research at the University of Oxford, OxSight pairs augmented reality and machine learning to help people with a residual level of sight, “offering a simplistic view of the world,” said Stephen Hicks, founder and head of innovation at OxSight. That lets people with reduced vision see shapes and movement, helping them to make their way around the world. Since everyone’s sight is different, OxSight uses machine learning to help customise the experience and learn the best way of displaying information.
Babylon Health
Babylon makes AI-driven chatbots, which were trialled by the NHS in north London to help triage patients, using the smart app as a digital 999 system. Patients enter their symptoms into the app, which analyses the data and asks for more information if necessary, helping to decide the seriousness of the illness or injury. The app isn’t for extreme emergencies, but rather the smaller queries that normally go to NHS 111, with Babylon claiming that the system takes an average of a dozen messages over a minute and a half to come to a decision.
Arterys
This medical imaging analysis tool is the first machine-learning app to be approved by the US Food and Drug Administration (FDA). Arterys learns as it goes, helping doctors examine heart images and data for diagnosis. To get FDA approval, it needed to post results on par with humans – and it does so in seconds; human experts can take as long as 30 minutes on the same task, the company said.