When Does an AI Update Become a Significant Change? - Regulatory Challenges for AI-based Medical Software under the MDR

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

AI-based medical devices in the European Union are regulated under the Medical Device Regulation (MDR), which is built on structured conformity assessment processes and clearly defined product states [1]. However, the continuous and multidimensional evolution of AI systems, including updates, is inherently misaligned with the binary classification of modifications into significant and nonsignificant changes as defined in current regulatory guidance [2]. To address this challenge, this paper proposes a perspective that interprets updates as transitions within a multidimensional performance space characterized by dimensions such as accuracy, robustness and data quality [3]. This perspective complements existing regulatory frameworks by enabling a more transparent and differentiated evaluation of system changes and may support more consistent, traceable, and lifecycle-oriented decision-making in conformity assessment processes, including within quality management systems and by Notified Bodies. This work focuses on static (“locked”) AI-based medical devices, where updates are deployed in a controlled manner.