FDA releases steps for building a review framework to authorize AI-based medical devices

FDA Commissioner Scott Gottlieb released a statement yesterday afternoon addressing the overgrowing presence of artificial intelligence and machine learning in healthcare delivery, noting several new steps it would like to take to develop a regulatory framework that could be used to promote the development of safe and effective medical devices that operate on advanced artificial intelligence (AI) algorithm software. Certain algorithms are now used to assist clinicians when they screen for certain diseases and they can also make treatment recommendations based on a rich variety of data. In addition to currently approved AI technologies, FDA says it expects to see an increasing amount of authorization activity as more medical device manufacturers include advanced AI algorithms in their products.

Right now, however, “the artificial intelligence technologies granted marketing authorization and cleared by the agency so far are generally called ‘locked’ algorithms that don’t continually adapt or learn every time the algorithm is used,” Gottlieb explained. “These locked algorithms are modified by the manufacturer at intervals, which includes training of the algorithm using new data, followed by manual verification and validation of the updated algorithm.”

However, as machine learning algorithms continue to evolve, they become adaptive or continuously learning algorithms that would not require having to make manual modifications to add learning and/or updates. Instead they digest the new user data placed in the algorithm from real-world use. Gottlieb pointed to breast cancer lesion detection powered by AI algorithms built into mammography equipment that would learn to increase in accuracy or identify specific subtypes of breast cancer based on real-time use and feedback. Details were released yesterday by FDA in a new white paper. 

According to FDA, the goal of the framework, in broad terms, is to:

·        assure that ongoing algorithm changes follow pre-specified performance objectives and change control plans

·         use a validation process that ensures improvements to the performance, safety and effectiveness of the artificial intelligence software

·         includes real-world monitoring of performance once the device is on the market to ensure safety and effectiveness are maintained

“We’re taking the first step toward developing a novel and tailored approach to help developers bring artificial intelligence devices to market by releasing a discussion paper,” Gottlieb announced. “Other steps in the future will include issuing draft guidance that will be informed by the input we receive.”

Gottlieb says FDA encourages diverse and thoughtful feedback on the paper from experts and stakeholders on the, adding that “as algorithms evolve, the FDA must also modernize our approach to regulating these products. We believe that guidance from the agency will help advance the development of these innovative products.