The present and future use of AI in healthcare

Current uses

While the use of AI in healthcare is expanding, "most of the actual, in-market, common uses of AI aren't super sexy," Smyth said.

One of the more common uses of AI and machine learning in healthcare is predicting the maintenance needs of medical equipment, Smythe said. It's not necessarily the most exciting use of AI, but it is important.

Using such tools and technologies at Philips, for example, "saves us a lot of money, saves customers a lot of money, saves downtime and makes it less likely a patient will show up for an examination and the machine isn't working, which has clinical ramifications, of course," Smythe said.

Over the next 25 years, we're going to be doing things we've dreamed about for 500 years.

AI and machine learning are also being used with clinical applications, such as medical imaging, where such technologies have made scanning for abnormalities in CT scans quicker and easier.

Roy Smythe, M.D.

For patients who need a series of CT scans over time, a "machine learning platform has learned which scan the clinician wants to see, which legion a clinician wants to see,'' and can even automatically measure the legion and locate past images of it for comparison, Smythe said.

Also, monitoring systems that use machine learning and AI can help predict if a patient is going into sepsis and needs immediate medical attention. The systems can also be used to take in massive amounts of data, which can later be used to enhance machine learning and deep learning algorithms.

Big technology companies like Philips are making steady advances with these systems and are starting to deploy the technologies steadily; so are startups, such as Israel-based Clew Medical.

Startups, too

Started in 2015, Clew has already formed partnerships with some major medical centers, including the Mayo Clinic in Rochester, Minn., and the Tel Aviv Sourasky Medical Center. These medical centers have provided Clew with data and have allowed the startup to run trials of its predictive clinical analytics system.

Built to integrate with a medical center's existing hardware and electronic health records (EHR) platform, Clew's system uses machine learning to take the output of monitoring devices and combine it with other data to look at multiple parameters together and to look for trends in clinical vectors, said Izik Itzhakov, vice president of business development at Clew.

Clew hopes the system will eventually provide medical centers with a prediction platform for clinicians and administrators.

"The main areas we started focusing on were septic shock, acute kidney injury and respiratory failure. However, we are working to look at other types of deteriorations, other clinical deteriorations, as well as administrative," Itzhakov said.

After successful early trials in which the platform's predictions were tested against what the physicians saw, Clew signed agreements with the UMass Memorial Health Care system in Massachusetts, WakeMed in Raleigh, N.C., and the Sheba Medical Center in Israel to continue testing the platform, this time allowing the system to alert clinicians in real-time to potential crises, Itzhakov explained.

Clew is working to take the technology further at Sheba Medical Center by having the platform run beyond just the critical care unit, he said.

Being able to identify sepsis in critical care is extremely helpful, Itzhakov said. "However, being able to tell if patients are deteriorating before they get to critical care is even more valuable."

Predicting predictions

Smythe, referring to Philips' own prediction systems, acknowledged that the platform, which is able to help predict sepsis, isn't "incredibly disease-specific now, but we're sort of easing into that as we collect more and more data."

The future of such prediction systems, he added, might not only help forecast potential issues and diseases, but it can also suggest what steps a patient should take and how medical staff should react.

Looking ahead, he said, the use of AI in healthcare could offer "sophisticated clinical decision support," by automatically adding patient data and finding the most effective treatment plan.

"Those data models take a ridiculous amount of data to train, and that's going to take some more time, but we'll get there," he said.

Machine learning and AI will also become commonplace in healthcare administration for tasks such as adding information to EHRs and billing patients, Smythe said.

Meanwhile, natural language processing tools could help physicians to easily create patient medical files by verbally discussing an examination with a patient as it is happening. After the exam is complete, an automated system can send a bill to the patient's smartphone immediately, Smythe said.

Democratization of healthcare

As these technologies advance, the use of AI in healthcare might eventually spur the democratization of healthcare -- making care more accessible to more people -- and it could help make care more affordable, Smythe said.

As AI tools advance, responsibility will be pushed down to "people lower and lower on the healthcare delivery food chain," he said. In the next few decades, that responsibility could be pushed all the way down to the patients themselves, as intelligent agents will be able to walk patients through a self-conducted exam for as much as a third of patient concerns.

Other AI-powered devices could help track an elderly person's movements and habits around their house to help evaluate their cognitive performance over time or to alert medical professionals if something seems wrong.

That may sound invasive to some, but Smythe said he believes "people are very accepting of these things ... because they want to age in place."

Before these potential future benefits of AI and machine learning are realized, however, Smyth said he believes the healthcare system will have to wade through a period of financial cost realignment and changing business models.

For the most part, healthcare systems in the U.S. have limited budgets and operate within less than 10% margins, Smyth said. The high costs of new AI and machine learning tools, among other outlays, could force systems to adapt their business models, with an increase in shared risk models between healthcare providers and vendors.

"There's been several times in history with technology where we've gone through these messy transition periods," Smythe said. "They don't feel good -- it feels clunky, and you feel that the technology is actually causing more problems than it's resolving.

"That's sort of the way it is in healthcare now, but we'll get past this and it will be a lot better."