Machine learning is coming to healthcare. From assisting with research to helping patients at the bedside, it is undoubtedly making its mark in the space.
Via email, Mayo Clinic CIO Christopher Ross corresponded with MedCity News to answer questions about the future of machine learning and when artificial intelligence will move past the toddler phase.
Behavioral Health, Interoperability and eConsent: Meeting the Demands of CMS Final Rule Compliance
In a webinar on April 16 at 1pm ET, Aneesh Chopra will moderate a discussion with executives from DocuSign, Velatura, and behavioral health providers on eConsent, health information exchange and compliance with the CMS Final Rule on interoperability.
Ross will deliver a keynote address, titled “Machine Learning, from Discovery to Action,” at MedCity INVEST on May 18 in Chicago.
What is Mayo Clinic doing in the healthcare AI space that’s unique and different from competitors?
I’m not sure if we’re doing much that’s unique and different. We have a number of projects underway, most of them early stage and exploratory, with a number of different partners. We have been working hard on Clinical Trials Matching with IBM Watson, and we’re encouraged by the results so far.
When is the last time you were blown away by what you saw machine learning could do?
A Deep-dive Into Specialty Pharma
A specialty drug is a class of prescription medications used to treat complex, chronic or rare medical conditions. Although this classification was originally intended to define the treatment of rare, also termed “orphan” diseases, affecting fewer than 200,000 people in the US, more recently, specialty drugs have emerged as the cornerstone of treatment for chronic and complex diseases such as cancer, autoimmune conditions, diabetes, hepatitis C, and HIV/AIDS.
I think we’ve all been impressed with examples outside healthcare, like autonomous vehicles and robotics. Watson winning Jeopardy was remarkable, and Google DeepMind mastering Go is similarly impressive. Consumer products like Amazon Echo and Google Home and Apple Siri have the potential to be effective in healthcare. When I get into my car, my phone increasingly offers route guidance that is highly accurate, even though I don’t drive to the same places at the same times. If the hyper-scale cloud companies can learn to predict and guide commutes for a couple billion people, one hopes that could be applied to various health management and maintenance tasks.
Specifically to healthcare, there’s been a number of recent published studies in which machine learning and various methods of discrimination have been applied to interpretation of radiology images and other complex domains. It’s pretty clear that machines will or have outpaced the ability of the human eye and brain to evaluate complex multi-dimensional types of data. Mayo has been working on natural language processing for many years, and the ability to extract real clinical meaning from sometimes imprecise text is getting better and better.
Earlier this year, you endorsed AI but also said it’s “like a two-year-old.” How long do you think it will be until AI moves out of the toddler phase?
Most AI today learns through a process called “supervised learning” in which an algorithm is exposed to many, many examples of a thing, and the algorithm adapts. If you show an image analysis algorithm a lot of pictures of cats, pretty soon it can distinguish a cat from a dog, and then a lion, and maybe Fluffy from Snowball. Facebook usually gets the tagging of images right, and Google’s photo tagging and sorting means I can actually find pictures I want. And this can be applied to hard things like radiological images for diagnosis. But that “supervised learning” sure reminds me of what it was like to chase a curious two-year-old around the house.
I can’t predict when AI adolescence will arrive. Right now supervised learning allows algorithms to build semantic models. So we can discriminate cats from lions, and we can relate the image of a cat to words: “cat” and “feline” and “felis catus.” One important next stage will occur when ontological models become involved. When most AI instances start, they have no ontological model of what a “doctor,” “patient” or “hospital” are, and how they relate to each other. Most AI uses mountains of data and millions of repetitions, after which an algorithm can create a kind of web of connections between ideas based on probabilities. What’s missing are ontological models, the kind of “mental models” we human beings carry around all the time in our heads. Once we understand the concept of a doctor and a patient, we can predict and generalize many of the things that a doctor and patient can do. Today, AI largely lacks that ability to abstract.
How can healthcare organizations work to ensure their focus is on the outcomes and not the technology?
Perhaps by staying focused on the goal. Mayo Clinic is focused on the needs of the patient, and the treatment of serious and complex conditions. If we aren’t advancing that, we are not doing the right things.
When — if ever — will the machines take over?
Well, when I’m in a plane, I’m grateful for our new overlords. Because each engine is streaming megabytes of data per minute to support predictive maintenance means engines don’t fail, and planes are less likely to be delayed for maintenance at the gate.
Machines have outpaced people at various discrete tasks for many generations. Manufacturing, retail, distribution and similar industries have already been changed forever by automation and AI. A decade ago the idea of an autonomous vehicle was science fiction. These technologies are in use in many industries, and healthcare shouldn’t ignore it. But I think healthcare is particularly complex, and depends more on human connection than almost any other human activity. So I’m optimistic we’ll see “augmented intelligence” where the computer and person pair to do better.
What is your vision for the future of machine learning in healthcare?
For now we see through a glass, darkly. Predictions or visions are unlikely to be very reliable at this stage. I think we all hope it will be one of several tools and investments which help improve health and treat disease.
Photo: monsitj, Getty Images
