FDA regulator examines AI's growing influence in cardiology

Navigating regulatory challenges for AI

Browning emphasized that the FDA continues to apply its existing regulatory frameworks, such as 510(k), premarket approval (PMA), and de novo classifications, while adapting them to meet the unique needs of AI technologies. 

According to this presentation, key challenges include:

  • Bias and data diversity: Ensuring datasets used in AI development are representative to avoid biased outcomes.
  • Algorithm transparency: Addressing the "black box" nature of AI, which can make it difficult for end users to understand how the system generates recommendations.
  • Iterative learning: Managing AI systems that evolve as new data becomes available, potentially requiring repeated evaluations. As the algorithms are improved with new iterations, companies often had to resubmit their AI for multiple FDA reviews.

To address these issues, the FDA has implemented innovative approaches, such as predetermined change control plans. These allow manufacturers to retrain and update their algorithms without submitting entirely new applications, provided they meet pre-established benchmarks.

"When the FDA is faced with a novel technology, it's very important for us to account for the challenges or the pitfalls those devices might fall into, as well as the benefits. And for AI algorithms, some of those challenges are the bias that could potentially be introduced if the data sets aren't sufficiently diverse and the opacity of the algorithms making it difficult to describe to the end user," Browning said.

AI is a different type of device technology, unlike what the FDA regulated prior to the first AI clearance in 1995. Browning said one challenge with these new technologies has been that the FDA did not get new authorities to regulate AI, so the agency has had to fit AI into the existing regulatory pathways. However, he said the agency has started using different methodologies to account for some of the challenges posed by AI.

Regulatory pathways and industry collaboration

Browning stressed the importance of a risk-based approach in regulating AI devices. Despite the novel nature of AI, the FDA still evaluates these products using the same fundamental questions:

  • Have the risks and benefits been sufficiently characterized?
  • Does the device meet safety and efficacy standards?
  • Are risks and benefits clearly communicated to users?

The FDA also leverages pre-submission pathways to collaborate with manufacturers early in the development process, aligning on validation strategies to streamline approvals.

Grappling with the increasing number of AI submissions

The exponential growth of AI in medicine, including cardiovascular care, continues to test the FDA’s resources. According to the FDA's website, the agency reviewed and cleared 250 new AI algorithms in the past year and the number submissions is increasing each year. However, Browning said the agency remains committed to fostering innovation while maintaining rigorous safety and efficacy standards.

"It's a challenge, but the industry is free to submit as many products as they're able to develop," Browning said.

Hear more insights from Browning in the video interview at the top of the page.