Impact of AI assistance on knee MRI reading time: A real-world multicenter study

Background Artificial intelligence (AI) has the potential to enhance radiology efficiency, but real-world data on its effect on reporting workflows is still limited.This study evaluates the impact of seamless AI integration on knee MRI reporting time in a real-world, multicenter radiology setting comparing standard reporting, partially integrated AI support, and fully AI-assisted structured reporting workflows. Methods We conducted a prospective, multi-center study across ten Swiss radiology centers from September 2023 to October 2024. Eight radiologists (four generalists, four MSK subspecialists) interpreted 1285 knee MRI exams over three sequential phases: (1) standard reporting without AI, (2) partial AI integration with image-only findings, and (3) full AI-assisted structured reporting with auto-populated templates. Reporting time was automatically recorded via the RIS radiology information system. Differences across phases were evaluated using the Kruskal–Wallis test and post-hoc Dunn’s tests. An inverse-variance weighted approach was used to compare mean reporting time differences. A post-study survey evaluated user satisfaction. Results Fully integrated AI-assisted reporting (Phase 3) significantly reduced average reporting time by 13.4 % (p < 0.01) compared to baseline. General radiologists benefited most (17.5 % reduction, p < 0.01), while MSK subspecialists experienced smaller improvements. Partial AI integration (Phase 2) did not reduce reporting time and occasionally increased it. User feedback indicated that 75 % found AI helpful and 62.5 % appreciated structured reporting integration. Conclusions Fully integrated AI-assisted structured reporting significantly reduced knee MRI interpretation time, especially for general radiologists. These findings support broader implementation of AI in routine musculoskeletal imaging workflows.