AAMI Roundtable on AI in Healthcare
By: Chris Hayhurst
June 16, 2021
Categories: Health Technology Management, HTM Professionals, Information Technology, Medical Device Manufacturers, Medical Device Manufacturing
For years, artificial intelligence promised to transform healthcare. So, how is AI being used in medical settings today, and what still needs to happen to make this vision a reality?
These were just some of the questions that were addressed in a recent AAMI roundtable featuring six experts from across the healthcare industry.
|[click to view]|
Fielding questions from AAMI editor-in-chief Gavin Stern, the participants shared their thoughts on everything from the biggest challenges facing healthcare-based AI to how the technology might improve patient safety.
“There’s a disconnect between what people expect and what is actually implemented today,” noted Jesse Ehrenfeld, MD, MPH, immediate past chair of the American Medical Association Board of Trustees and a professor of anesthesiology at the Medical College of Wisconsin. As a clinician, he said, he can imagine “lots of ways” that AI might be deployed in his work with patients. So far, though—in his experience—the technology has mostly come up short.
“There’s a desire and a need,” Ehrenfeld said, “because we all experience the power of AI in our daily lives as consumers walking around America, and yet it hasn’t really been brought into healthcare in a meaningful kind of way.”
Joining Stern and Ehrenfeld in the discussion were Emily Hoefer, director of standards at AAMI; Zack Hornberger, director of cybersecurity and informatics at the Medical Imaging & Technology Alliance (MITA); Emily Patterson, PhD, a human factors engineering expert and professor at Ohio State University; Mehdi Farhangi, a computer scientist with the Division of Imaging, Diagnostics, and Software Reliability at the Food and Drug Administration (FDA); and Pat Baird, regulatory head of global software standards at Philips.
Baird, who along with Ehrenfeld is the current co-chair of a new AAMI standards committee focused on AI, agreed with his colleague when he opened the discussion with an observation.
“Remember the old quote, ‘If you build it, they will come?’ Well, I’ve noticed that that really doesn’t work for healthcare.” When it comes to AI, Baird added, “one of the challenges we have is figuring out how to fit [it] into the existing healthcare ecosystem.” That includes addressing things like standards and regulations, but it also includes everyday issues like how to bill for AI-assisted services and questions around legal liability. “If something goes wrong, who is it that’s going to get sued?” he asked.
On Algorithms, Accuracy, and Areas of Advancement
While the roundtable participants spent much of their hour together parsing the hurdles associated with AI, they also expressed optimism that the technology is positioned to improve healthcare in many ways. Patterson, for example, talked about AI algorithms that could identify fetuses that were at high risk of sepsis at birth, or identify when a patient was likely to have cancer. Such tools could prove promising, she said, but only if the information they provide is presented, explained, and “contextualized by a physician.”
Another area where AI can make a difference is in image recognition and improving clinical workflows, Patterson said. If an algorithm can identify with a high degree of accuracy the one or two images in a stack of hundreds most likely to show a cancerous tumor, for example, that could help the pathologist determine which images to evaluate first. Farhangi, for his part, agreed with Patterson. When used as a secondary or concurrent means of diagnosis, AI can help bring “suspicious cases” to the surface, he said, and that should be of benefit to busy clinicians.
While many in healthcare find artificial intelligence to be something of a mystery, Baird said he believes the industry is ready for the technology—we just have to start thinking about it in a different way.
“Some people think you need totally new regulations, totally new approaches to everything,” but that’s not necessarily the case, he said. “When I started digging in, and I started looking at what’s important for a quality AI product versus what’s important to other kinds of things, I found a lot of stuff that we already know how to do.” One could worry about the quality of data that is used to train healthcare’s AI algorithms, for example, or they could simply address those concerns by applying the same processes they follow every day.
In healthcare technology management, Baird explained, BMETs are always testing and adjusting equipment based on their findings. “Well, in a way, isn’t that what we’re talking about here” with AI? “The process isn’t 100 percent the same, but at least it’s a starting point—we’re not staring at a blank page. So let’s take and reuse [what] we already know how to do, and adapt as necessary.”