Twenty years ago, in 1998, you would have been hard-pressed to find a single hospital room with a personal computer in it.
Patient files were kept in filing cabinets. Prescriptions were written by hand. The Human Genome Project was still just halfway through a years-long, multi-billion-dollar effort to sequence the DNA of the human race. In short, there wasn’t a huge abundance of data on our health.
Today, electronic medical records and advanced genetic sequencing have completely changed the landscape — and brought challenges and opportunities that are almost impossible for humans to tackle on their own. That’s where artificial intelligence steps in.
AI is finding fertile ground for growth in hospitals and medical labs around the world, promising to give human health a boost as it addresses everything from preventing heart attacks to revolutionizing how we diagnose diseases. That interplay is the topic of our most recent Health Tech Podcast, and if the experts are to be believed, it’s just the beginning.
“I see this possibility of precision health, where people are the most fundamental thing in the Internet of Things,” said Peter Lee, a corporate vice president at Microsoft who leads the company’s NExT program. “When we’re looking 10 years out, that sort of precision in diagnosis and treatment, I think, can be incredibly powerful.”
Lee works with industry partners to implement cutting-edge technologies, including AI, and his team started looking into health applications of AI about a year ago.
“Honestly, that feels a little bit like being thrown into the middle of the Pacific Ocean and asked to find land, because healthcare is just such a huge, huge space,” Lee said. “But as time has gone on, over the past year, we’ve really gotten completely sucked into it and we’re pretty excited.”
Ankur Teredesai, a longtime University of Washington data scientist and co-founder and CTO of health AI startup KenSci, started studying artificial intelligence twenty years ago, back in 1998. At that point, the technology was still in its infancy, much as health data was at the time.
“Just this abstract concept of intelligence, which could be derived from a computer program, was fascinating to me,” Teredesai said. He went on to found the Center for Data Science at UW Tacoma in 2010 and saw a huge opportunity in health.
“There was a ready availability of data emerging” in health, he said, “but there were hardly any data scientists that were looking at solving big problems in this space.”
KenSci now works on AI that predicts which patients will get sick and helps hospitals intervene early.
The confluence of huge amounts of data, advancing AI technology and the many challenges facing healthcare in the U.S. puts the industry in a unique space, ready for a new way of solving problems.
One issue Microsoft NExT is taking on is personalized medicine, treatments or other health actions that are tailored to an individual based on genes and other health data.
Lee said precision medicine is interesting to Microsoft for two reasons.
“One is: Precision medicine still depends — a lot — on fundamental research and especially research in AI and machine learning,” he said, “and two, the computing workloads are really very data dependent and typically involve very large volumes of data.”
Microsoft just announced a new, multi-hundred million dollar precision medicine project with Adaptive Biotechnologies, a Seattle company that specializes in sequencing the genes of immune cells. Microsoft will help Adaptive build a machine learning program that can scan those genes and use the data to figure out what diseases someone might have, potentially months or years before they show symptoms.
Microsoft is also working on a patient-facing AI chatbot that will help people navigate their healthcare and insurance benefits — and although those two projects seem totally different on their face, Lee says the technology behind them is actually very similar.
Introducing GeekWire’s Health Tech podcast, exploring the frontiers of digital health
“In a way, both the health bot technology and what we’re doing with Adaptive — from an AI perspective — have common roots in what we do today in machine learning for language processing,” he said.
The health bot is rooted in language because it needs to have a natural language interface that can hold a conversation with a user. It turns out the biotech project with Adaptive is also a language problem.
“You have these antigens that are indicative of some disease state in your body. Those antigens are like words that are telling some story about what’s going on in your body. The T-cell receptors that are part of your immune system are like a translation of those words into a new language,” Lee explained. Adaptive’s technology lets them read and understand the genetics of a T-cell.
“From an AI perspective, what we’re trying to do is use machine learning to do the language translation from the T-cell receptors back to the antigens so that we can understand what your body is saying,” Lee said.
Teredesai and KenSci are working on a different kind of artificial intelligence problem: Predicting future events given a past history.
KenSci’s technology uses patient data from electronic medical records to do a few things, chief among them predicting which patients will get chronic diseases.
“A chronic condition patient — a patient who is a diabetic, or has an episode of heart failure — often starts off as a patient that is normal,” Teredesai said.
“They are able to take care of themselves, they are [at] very low risk of mortality. And it is the small details in their daily lives that — if they manage properly — they can have a very, very successful life that is pain-free, disease free and leads to a desirable outcome where mortality can be managed, to a great extent,” he said.
KenSci also works on tools that help hospitals see patterns in how patients are faring. It might help a hospital change certain policies or give caregivers new training to improve health outcomes, for example.
Teredesai also emphasized that KenSci’s products — and all AI programs — are not replacements for humans in the health system. He actually prefers to call AI “augmented intelligence” or “assistive intelligence,” making the point that it must work in tandem with doctors, “rather than act like death robots that are controlling the entire healthcare ecosystem,” he said.
Lee also raised a point of concern about AI in the health field, namely its reliance on correlations that make up a pattern.
“Medicine is, properly, not based on correlations, but is based on causal relationships — and it needs to be that way. That’s why medical research is really founded on ideas about having controlled experiments, about really understanding statistical significance and really being very wary of making decisions based only on statistical correlations,” Lee said.
“So there’s a gap right now between what we are actually doing today in the world with machine learning and AI and what medical science has always been based on,” he said. That gap must be closed if AI is to reach its full potential in the health world.
And Lee predicts a bold and successful future for AI in health. At the moment, he says, AI is helping build one-off clinical tools that are useful in a certain situation.
“What I see coming after that is something that really would be enabled by a much broader look across large populations, to look at longitudinal patient records for millions or even hundreds of millions of people and understand them in ways that can be actionable by clinicians and by healthcare organizations,” he said.
“If you were looking 10 years out, I see this possibility of precision health where people are the most fundamental thing in the Internet of Things,” he said. “It’s not just your Fitbit but it’s your genome, it’s your activities all day every day.”
“It’s where you live, it’s who is living around you, what you’re eating… those things are creating a kind of digital avatar that can virtually see an intelligent doctor. Maybe every day, or even every hour.”
He compared the idea to the way the health of cars is monitored today: If something goes wrong in the car, sensors discover it and notify the driver before he or she can even tell there’s a problem. The same could be done with different health monitoring systems for someone’s body.
“Today, health care is 95 percent about people and chemistry and drugs and so on and 5 percent compute, and that gets to a world where health care really starts to flip. It will approach something more like 5 percent that stuff and 95 percent compute,” Lee said.
It may be a decade or two before we carry virtual, AI doctors in our pockets, but it’s clear AI is already having a stark impact on the health world.