Delays between CE mark and FDA regulatory approval of AI-enabled software for radiology

Abstract

Artificial intelligence (AI)-enabled radiology software requires regulatory authorisation before marketing in Europe (EU) and the United States, but cross-jurisdictional approval sequencing remains poorly characterised. We analysed 239 AI-enabled radiology software devices with CE marking and/or FDA clearance from a publicly available European database (Health AI Register) and FDA sources using Kaplan–Meier methods and Cox models. Overall, 128 devices were CE-marked only, 95 were CE-marked before FDA clearance, and 16 were FDA-cleared before CE marking. Among dual-authorised devices, CE-first devices had longer median lags to second authorisation than FDA-first devices (17.5 months [IQR, 8.0–35.5] versus 3.5 [IQR, 1.0–11.3]; p  < 0.001). Radiograph interpretation software was associated with longer time to second authorisation (hazard ratio, 0.42; 95% CI, 0.21–0.84; p  = 0.01), while EU Class IIa classification (4.99; 1.77–14.04; p  < 0.01) and peer-reviewed evidence (1.44; 0.83–2.49; p  = 0.19) showed shorter intervals. These findings highlight persistent transatlantic asymmetry and underscore the need for greater regulatory coordination and transparency.