Hundreds of approved devices use artificial intelligence to help physicians diagnose patients faster and more accurately. Brooke & Associates is a legal and regulatory advisory firm that helps medical device makers get AI-powered devices through FDA pre-market approval.

In Episode #40 of the MedTech Speed to Data podcast, Key Tech’s Andy Rogers and Lei Zong speak with the firm’s managing member, Jason Brooke, about the FDA’s latest guidance to medical device developers for integrating AI into their products.

Need to know

• AI’s role in MedTech — AI identifies otherwise undetectable data patterns that humans can apply in clinically meaningful ways.
• FDA’s AI staffing surges — The agency accelerated hiring to develop internal AI applications and support pre-market reviews of new AI-powered devices.
• Radiological imaging leads the pack — More than half of 900+ FDA-approved AI-based products are in radiological imaging.
• Other fields are catching up — Cardiology and neurology applications are more recent entrants in AI-powered devices, but their numbers are growing.

The nitty-gritty

The FDA published “Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations” in early 2025 to explain how it will address AI’s adaptive nature in medical device regulation.

“This guidance is really focused on a total product lifecycle approach,” Brooke explains.

Good management practices govern traditional medical device development, so documenting the development process in pre-market submissions is not as critical. AI model development is different because the model can evolve once in service.

“There’s a level of information that’s necessary in submissions for AI-based technologies that we haven’t had to provide to the FDA before,” Brooke says. “They want a lot of information,” Brooke says. “That’s an area I think may be problematic because a lot of that is somewhat trade secret.”

AI-specific guidance touches almost every aspect of a company’s submission, from risk assessment to labeling to cybersecurity. Brooke highlighted how the FDA’s approach to AI data management could change development practices to ensure the independence of training and validation data sets. For example, companies must separate their clinical sites geographically and temporally.

“This guidance gets into the weeds,” Brooke says. “It’s important for companies to understand this if they’re developing an AI-based product.”

Data that made the difference:

In addition to discussing the FDA’s proposed AI regulations, Brooke discusses the challenges companies face in bringing AI-powered medical devices to market.

“If you take away anything from this podcast,” Brooke says, “it’s that there’s a lot of burden associated with developing an AI-based medical device. If you don’t need to, then I wouldn’t recommend doing it.”

Slow and steady wins the race. Do your homework, plan for the FDA review, and then engage the agency at the right time to get them on board.

Thoroughly characterize your data sources. Devices like ECGs can vary by vendor, model, site location, patient, and many other factors. The FDA wants to know how this variability could affect the downstream AI model.

Develop a strong clinical validation plan. The FDA will limit claims and require disclosures when a device that performs well overall underperforms among certain patient groups.