By Richard Dargan
Direct integration of AI-generated findings into structured knee MRI reports significantly reduces reporting time, especially for general radiologists, according to research presented Monday.
Recent years have seen a proliferation of AI tools promising to improve workflow efficiency. The benefits of these tools, however, are unclear.
“Time gain while using AI tools for reporting is controversial, as AI results are typically displayed as a separate series in the PACS and might add an additional step in the radiologist workflow,” said study lead author Benoit Rizk, MD, a radiologist in private practice in Villars-sur Glane, Switzerland. “We wanted to know if the efficiency in reporting time for these tools was real and identify who would most benefit from it: general radiologists or subspecialized radiologists.”
Dr. Rizk and colleagues in Switzerland, where knee MRIs are especially common due to the prevalence of skiing-related injuries, evaluated the impact of AI integration on knee MRI reporting time across 10 private radiology centers.
The researchers used a knee MRI AI tool that provides several outputs describing the state of the anterior cruciate ligament and other important structures in the knee such as the cartilage, menisci and medial collateral ligament.
“During acquisition, the technologist begins with the two sequences necessary for AI analysis, so usually the AI output is available at the end of the examination when the radiologist wants to report it,” Dr. Rizk said. “When the radiologist opens the standardized structured report template to report the exam in the radiology information system, the different AI outputs prefill the report in the description and in the impression.”
“For general radiologists, our study showed a significant time gain in reporting time with prefilled reports compared to standard reporting without AI. In a context of increasing workload, this helps to cope with the number of daily cases to read, as knee MRIs are frequent, and to decrease the cognitive workload of radiologists.”
Benoit Rizk, MD
The researchers allocated 1,285 knee MRI exams across three reporting workflows: standard reporting without AI, reporting with AI results available in the PACS, and reporting with AI outputs prefilling structured standardized report templates.
Eight radiologists, including four musculoskeletal subspecialists and four generalists, participated in the study.
Fully integrated AI-assisted reporting reduced overall knee MRI reporting time by 13.4% compared to standard reporting. General radiologists saw a 17.5% decrease. Musculoskeletal specialists also benefited, albeit to a lesser extent, thanks to already optimized workflows, according to Dr. Rizk. “Partial AI integration did not yield time savings and, in some cases, slightly increased reporting time, likely due to workflow modifications,” he said.
“For general radiologists, our study showed a significant time gain in reporting time with prefilled reports compared to standard reporting without AI,” Dr. Rizk said. “In a context of increasing workload, this helps to cope with the number of daily cases to read, as knee MRIs are frequent, and to decrease the cognitive workload of radiologists.”
Qualitative feedback revealed that 75% of radiologists felt AI facilitated their work, and 62.5% found structured report integration beneficial, particularly for straightforward cases.
The researchers are working to enhance the algorithm's capabilities using foundational models to broaden the scope of its outputs and improve precision.
They are also developing more prefilled, standardized, structured report templates with AI outputs, including for chest CT, bone age, lumbar spine MRI and musculoskeletal measurements on radiographs.
Study coauthors are Federica Zanca, PhD; Philippine Cordelle; Benoit Dufour; Natalie Heracleous, PhD; Cyril Thouly; and Sergey Morozov, PhD, MPH.
Access the session, “Impact of AI Assistance on Knee MRI Reading Time: A Real-World Multicenter Study,” (M6-SSMK03-4) on demand at RSNA.org/MeetingCentral.
© 2025 RSNA.
The RSNA 2025 Daily Bulletin is the official publication of the 110th Scientific Assembly and Annual Meeting of the Radiological Society of North America. Published online Sunday, November 30 — Thursday, December 4.
The RSNA 2025 Daily Bulletin is owned and published by the Radiological Society of North America, Inc., 820 Jorie Blvd., Suite 200, Oak Brook, IL 60523.
