A semi-automated natural language processing-based analytical pipeline for identifying structural shifts in MRI safety incidents: an analysis of FDA MAUDE reports (2018–2024)

Abstract Despite the widespread adoption of magnetic resonance imaging (MRI), safety incidents during MRI examinations continue to be reported. Building upon prior FDA-led analyses of MAUDE data, this study extends the temporal scope and analytical depth to systematically examine MRI-related adverse events reported between 2018 and 2024. Using a semi-automated classification framework that combined keyword-based text analysis with expert review, MRI-related incidents were categorized into granular safety domains, and inter-category correlations and associations were analyzed to characterize incident patterns at the level of individual events. The results demonstrate that MRI-related incidents persist annually, with no substantial reduction observed in Operation- and RF/Gradient-related events, which together account for a large proportion of reported cases. In parallel, implant-related incidents have become increasingly prominent and are associated with the majority of reported MRI events, extending beyond any single failure category. Although further subclassification of implant-related incidents was attempted, substantial semantic overlap among incident descriptions highlighted the complex and intertwined nature of implant-associated MRI safety challenges. Overall, these findings indicate that contemporary MRI safety risks are increasingly shaped by interactions among implants, software-based systems, and operational factors, rather than isolated hardware failures alone. Furthermore, the developed analytical pipeline establishes a reproducible measurement framework for large-scale medical device surveillance. This study provides data-driven insights that may inform future medical device design, more conservative MRI safety guidelines, and expanded education for both medical professionals and patients.