The Future of Artificial Intelligence in Orthopedics

Siri, Alexa, Tesla, Amazon and Netflix.

All use Artificial Intelligence (AI) to think, learn and then converse, entertain, educate and even drive our cars.

Inescapably, as it weaves its way into our personal lives, AI will transform orthopedic care. Are we ready? Are our patients ready?

At a recent Boston artificial intelligence conference (The World Medical Innovation Forum) that subject was tackled by AI theoreticians and AI entrepreneurs from China, the U.S. and Canada.

Here is a summary of what we learned.

What, Exactly, Is AI?

Artificial Intelligence (AI) is “intelligence” exhibited by machines.

Weak AI, the form of AI where programs are developed to perform specific tasks, is probably the dominant form being used today. You see it at work in diagnostics software, in the controls for robotic assist devices and in these new, smart remote sensing programs.

A machine imbued with AI, like a semi-autonomous car or a surgical assist robot, is defined as a machine that can: 1) perceive its environment and, 2) take actions based on those perceptions to maximize functional performance or achieve a goal.

AI Is Ripe With Hype

One fact to keep in mind, is that the ACTUAL software behind today’s AI land rush was created a decade or more ago.

Two recurrent phrases we heard at the Boston AI Forum were: “AI is Whatever Hasn’t Been Done Yet” and “Garbage in, Garbage out.”

Billions of dollars are being thrown at AI and much of it, we think, is chasing shiny objects. Also, we listened to dozens of talks by top academic medical researchers who pitched their registry data as a valuable database for AI application.

No doubt, AI will be the new buzz word in hundreds of National Institutes of Health (NIH) grant applications.

Having raised the issues of hype and flawed data, it is also true, we think, that taking many shots on goal is a winning Ai investment strategy. This market is so new, so large that even if 8 out of 10 investments fail, the 2 winners will make up the difference.

Today’s AI land grab is about obtaining proprietary control of massive datasets for which only AI can unlock economic value. Big Data can be an AI oil field. And healthcare may be the next Saudi Arabia.

But not every massive data set has economic value.

There Are Three Basic Forms of AI

These are the three main types of AI software:

Analytical:

Analytical AI is software which learns from past experience. It is a form of rule-based software. For example, an analytical AI program could learn each of 10,000 steps for a hip replacement (THA) operation based on a single, expert surgeon. That establishes the “rules” for a THA surgery. The analytical software then uses those rules to generate a cognitive representation of that surgery. It then tests that representation of THA against thousands of actual operations—analyze the differences, and thereby “learn” and adapt from that experience.

Human-inspired:

Human-inspired AI tries to mimic the human brain’s method of processing information and also uses elements from cognitive and emotional human intelligence to understand human emotions and consider them in their decision making.

Humanized artificial intelligence.

Finally, the most ambitious AI software attempts to elicit characteristics of all types of human competencies (i.e., cognitive, emotional, and social intelligence), is able to be self-conscious and is self-aware in interactions with others. Humanized AI doesn’t exist in reality, yet. In Hollywood, sure.

The first concept to keep in mind is that AI is still a rudimentary product. It will almost certainly become better at reasoning, knowledge representation, planning, learning, natural language processing, perception and the ability to move and manipulate objects.

The building blocks of AI are well known, powerful and entirely capable of taking AI through this evolution to cognition.

Those building blocks are statistical methods, computational intelligence, search and mathematical optimization, artificial neural networks, and combinatory mash-ups of statistics, probability and economics.

AI’s Guttenberg Moment – Chess and Go

Beating world champion Chess and Go Masters became “Guttenberg” moments in the history of AI. In both cases, AI computers demonstrated near super-human reasoning and cognitive intelligence capabilities.

In the Chess example, IBM’s Deep Blue, using a rule based AI software system, beat world champion Garry Kasparov in 1997. It did so by brute computing power. It could respond to every Kasparovian move by analyzing future potential moves and assign values at super-human speeds.

In the case of Go, the ancient Chinese board game which has infinitely more possible moves than chess, a different type of AI program was employed to beat the reigning world champion in 2016. That type of AI program is based on Artificial Neural Networks (ANNs).

Artificial neural networks (ANNs) learn more (or self-teach) by analyzing and comparing new data input. For example, an ANN can learn to identify images that contain cats by analyzing example images that have been manually labeled as “cat” or “no cat” and using the analytic results to identify cats in other images.

Go, the ancient Chinese board game has so many possible moves that it puts rule-based AI algorithms into computer hell. There are 361 possible first moves in GO. And each subsequent move also has 361 permutations. Winners don’t know if they’ve won until the game is over. In short, Go requires players to play at an abstract and intuitive level—as opposed to a rule-based approach.

The only way to be a consistent winner at Go is to be very, very good at pattern recognition.

An ANN designed software program called AlphaGo used a “policy neural network” AI algorithm to establish values for each move and then it calculated the value of each pattern of Go pieces in terms of a winning outcome. AlphaGo then played against itself millions of times in order to create patterns of winning at Go.

As it generated these patterns of Go, it recalibrated the values for each pattern. In short, it learned as it played. It used its neural network to become better. Until, literally, it became super-human.

AlphaGo played 60 professional games on a Go website under the name “Master” and won every single game, against dozens of world champions. In 2016 it crushed the best Go player in the world.

That event was China’s “Sputnik” moment and is largely credited with fueling a massive wave of funding for AI companies and research in China.

Applying AI to Orthopedics

Accepting, first, that AI is already in orthopedics in the form of diagnostic programs, robot assist programs and patient treatment algorithms, the question is where (and when) will AI deliver its most impact effect on orthopedic physicians and patients.

Based on the presentations at the Boston conference, one application stood out above all others—applying AI to electronic health records.

Electronic health records (EHR) are mixed blessing for physicians in the U.S. While the data is nice, the cost is rising levels of burn-out among physicians. This not a trivial problem.

One speaker, who runs the data department at a major hospital network, described her hospital’s interest in using AI to “read and hear” interactions between the physician and patient and “write” the electronic health record based on the recording of those interaction. Essentially, she described using an advanced rule-based AI algorithm to compare tens of thousands of recorded interactions to, in classic analytical AI form, create an autonomous and highly accurate EHR writing program.

That, is seems to me, would be a very welcome tool for all physicians.

The other start-up company that created serious buzz at the Boston Medical Innovation Forum was Yidu Cloud, a Chinese electronic health records aggregator.

Yidu Cloud

Chinese hospitals run software from approximately 4,000 vendors. By contrast, the U.S. has just a handful of major EHR vendors. Even within Chinese hospitals, there is no consistency and a patient can have as many as 40 different ID numbers—each from a different department.

Yidu Cloud was founded to make sense of this Tower of Babel data.

In its first three years, the company processed records from over 300 million patients by setting up cloud services in more than 700 hospitals.

Yidu also works with medical research institutes to form partnerships with hospitals to use the underlying Yidu controlled data set to research disease and treatments.

Yidu co-founder, Gong Rujing, was a speaker at the Boston conference and said, “It’s extremely difficult to make the data from unusable to usable.” She employs, for example, more than 200 physicians who perform a quality control function. When a patient’s record doesn’t meet Yidu’s quality control (QC) standard, the company sends it back to the originating hospital for review.

Yidu received incredible amounts of funding from both the Chinese Government, which hopes to establish three large national databases by 2020, and private investors. (the U.S. government under Trump has cut AI funding).

Yidu has spent more than $100 million on data collection and analysis. Said Rujing, “Each of the hospitals we work with is huge. Each has 50 to 200 systems with up to 20,000 outpatients per day. We deal with billions of data records.”

Processing that volume of data pushed Yidu to create a machine learning approach which they call “named-entity recognition” or NER.

In action, Yidu sets up a private cloud system within each hospital and provides access to a joint cloud. Servers exist at the hospital, not Yidu. The hospitals then authorize Yidu to extract the data and the hospital provides data channels for that purpose. In effect, the hospital uploads their data to Yidu.

Yidu then cleans the data using its NER algorithms, an AI natural language processing program and does all this while keeping individual patient identity anonymous.

Yidu has yet to generate a single dollar of revenue! Investors, which include the government of China, understand that this is a form of land grab and view this as a strategic investment.

What’s Next?

The AI phenomenon looks very much like the early days of the Internet.

Medicine generally, and orthopedics specifically, is an outstanding application for AI. The trillions of dollars, millions of healthcare providers and billions of patients which comprise the global medical industrial complex could benefit significantly from these tools.

Getting from here to there, however, will be an exciting ride for us all.

If we were to guess, we would say that the early winners in orthopedic AI will be those companies who tackle the most practical but complex problems like electronic health records.