The lucent yet opaque challenge of regulating artificial intelligence in radiology
James Hillis, Jacob J. Visser, Edward R. Scheffer Cliff, Kelly van der Geest – Aspers, Bernardo C. Bizzo, Keith J. Dreyer, Jeremias Adams‐Prassl, Katherine P. Andriole
Abstract
The potential applications of artificial intelligence and machine learning (AI/ML) in medicine are progressing rapidly. AI is a broad term that refers to the intelligence of computer and software systems, while ML is a type of AI involving computers learning through pattern recognition methods including artificial neural networks. Radiology is a frontrunner in this space: between 2015 and early 2020, 129 radiology AI/ML devices received regulatory clearance from the United States Food and Drug Administration (FDA), and 126 devices received the Conformité Européenne (CE) mark in Europe 1 . These approvals are only accelerating, with the FDA clearing 126 radiology AI/ML devices in the twelve months to July 2022 2 . Both the speed and volume of AI/ML devices present a delicate balance for regulatory bodies: ensuring the safety and effectiveness of devices while keeping pace with the clinical innovation and value that they may provide.