We Need More Than Just Data—We Need Knowledge

When it comes to using AI in the medical device industry, Pat Baird, senior regulatory specialist at Philips, believes that understanding and accounting for potential biases in the data is one of the most important aspects to consider.

Susan Shepard

July 10, 2024

4 Min Read
AI in medical devices Pat Baird Advanced Manufacturing Keynote
Toowongsa Anurak/iStock/Getty Images Plus via Getty Images

At a Glance

  • Baird will deliver the keynote, “The Next Iteration of AI in Medical Devices,” at Advanced Manufacturing Minneapolis.
  • The event features the following shows: MD&M Minneapolis, D&M Minneapolis, ATX Minneapolis, MinnPack, & Plastec Minneapolis.

Regulating AI in medical devices is “interesting,” said Pat Baird, senior regulatory specialist at Philips, because the regulators themselves are still trying to figure out what they should be doing. “They are still trying to feel their way through it,” he said in an interview with Design News. “But they do have a couple projects to sort of establish the foundation for future stuff.”

In his October 16 keynote address at Advanced Manufacturing Minneapolis, “The Next Iteration of AI in Medical Devices,” Baird will explore several aspects of AI in medical devices. As the cochair of one of the artificial intelligence committees for the International Medical Device Regulators Forum (IMDRF), he thinks one of the biggest issues surrounding AI for regulators is in the data that it is fed. He compared it with using a bad batch of raw materials on a manufacturing line. “The product that you’re making for the next couple of months isn’t going to be as good because the supplier screwed up something,” he explained. “For AI, [that something] ends up being the data that it uses to train and test it.”

Baird believes that there are a lot of supply-chain types of issues that can be leveraged in regulating AI. “You don't think of patient records as coming from a supplier, but I think that metaphor works.”

AI-in-medical-devices-Pat-Baird-headshot.jpg

Dealing With Potential Bias in Data

Maybe one of the biggest concerns for regulators is bias in its “supply chain” or data, Baird said. To illustrate this, he told the story of a friend who was looking at data of people with rare diseases in a particular area. She found that there was one hospital where no one was given an expensive cancer treatment. “So, the people in that zip code must be really healthy because there’s never any expensive cancer treatments?” Baird said. “Well, no, it’s because the people are so poor they can’t afford the treatment,” he explained. “That's why you don't see medical records and billing of chemotherapy there. It's not because they're healthier. It's because they can't afford it.

“There's this deeper level of understanding that we have to have about our patients, about how healthcare really works,” Baird continued. “We need more than just data. We need knowledge.”

Going forward, Baird said he thinks manufacturers will have to prove to regulators why they think their data is good and accurately represents their population. “I think that’s really sort of the first baby step,” he said. “Do we know where we are now, do we know what biases that we have, do we know what patient populations this doesn't really apply to because we don't have any data from them?” he asked.

FDA does have a draft guidance and is very close to publishing their final version of it, he noted. One of the most notable items it includes is the Predetermined Change Control Plan (PCCP). “After your product is launched, you’re going to have a lot more data than you did when you first were doing your research for it,” Baird said. “You’re going to have more patients and more data to improve your product,” he explained. “FDA really likes this idea and is pushing companies to update their products and update it as often as they’re getting more data because it’s going to improve performance.”

Predictive AI Versus Generative AI

Baird also spoke with Design News about the differences between predictive and generative AI. Predictive AI, he said, can look at a picture of a cat, and then tell the user that the image is a cat. In the healthcare industry, if you take MRI and CT scan images, there is a lot of data there that doctors have labeled as being cancer and also noncancerous, he said. “Then you just feed it to the AI and then it can do the screening. It can start predicting,” he explained. “So that's one application, helping them diagnose things.”

But, Baird said, generative AI creates new things. “You feed it data, and it will come up with a summary for the patients who don’t know the medical jargon. They can explain it in patient terms, what the disease is, what the therapy is going to be,” he said.  

“Classify versus create,” Baird summed it up.

Baird will delve further into the latest AI regulations and how they relate to medical device design, and what attendees can expect to see in terms of AI developments in the next few years in the October 16 keynote, “The Next Iteration of AI in Medical Devices,” at Advanced Manufacturing Minneapolis. The session will be held 10:15 to 11:00 a.m. in the Engineering Theater. 

Also delivering a keynote will be Amy Alexander, unit head, mechanical development & applied computational engineering, division of engineering, Mayo Clinic. She will present "Printing a Healthier Tomorrow: How 3D Technology is Shaping the Future of Medicine" on October 17 at 10:15 to 11:00 a.m. in the Engineering Theater. 

Advanced Manufacturing Minneapolis features the following co-located shows: MD&M Minneapolis, Design & Manufacturing Minneapolis, ATX Minneapolis, MinnPack, and Plastec Minneapolis. The event will be held October 16-17 at the Minneapolis Convention Center.

About the Author(s)

Susan Shepard

Susan Shepard is a freelance contributor to Design News and MD+DI.

Sign up for the Design News Daily newsletter.

You May Also Like