Impact of Deep-Learning Reconstruction on MRI Workflows: A Retrospective Analysis at a Large Academic Tertiary Center.

PURPOSE: Artificial intelligence is increasingly integrated in clinical practice. In radiological imaging, deep-learning (DL)-based image reconstruction techniques show potential for accelerating and enhancing the quality of examination procedures in magnetic resonance imaging (MRI). This study evaluated the impact of DL on MRI workflows and protocols over a one-year period following clinical implementation. MATERIALS AND METHODS: This retrospective, single-center study included 8,183 MRI examinations performed between 2023 and 2024 to assess daily examination volume, examination duration, and rates of repeated and aborted scans. Furthermore, 43 comparable sequences and 34 protocols on a 1.5T MRI (Siemens Magnetom Sola) were analyzed and surveys were conducted among 23 medical staff members to evaluate the perceived effects of DL on workflow efficiency, diagnostic performance, stress levels, and technology acceptance. RESULTS: 88% of the DL-reconstructed sequences demonstrated reduced acquisition times, yielding an overall 13% reduction in protocol duration, despite higher resolution or reduced slice thickness. On days without anesthesia support, the average examination time decreased by 11%, accompanied by increased patient throughput (+7.2%) and fewer repeats (-25%) and aborts (-100%). Medical staff reported high levels of technology acceptance (90%), perceived improvements in image quality (90.5%), and reduced stress levels. Shorter times required for generating medical reports were noted by 45.5% of respondents, particularly among residents (70%). CONCLUSION: DL enables significant potential for MRI-workflow optimization and protocol improvements, while maintaining high staff satisfaction, thereby highlighting the great potential of DL in radiology practice. KEY POINTS: · While maintaining or increasing image quality, DL reduced examination duration (-11%) and protocol length (-13%).. · Patient throughput increased (+7.2%), while repeats (-25%) decreased.. · Staff reported high technology acceptance, lower stress and faster report generation, particularly residents.. CITATION FORMAT: · Rizzetti M, Michael AE, Almansour H et al. Impact of Deep-Learning Reconstruction on MRI Workflows: A Retrospective Analysis at a Large Academic Tertiary Center. Rofo 2026; DOI 10.1055/a-2857-0974.