By Lynn Antonopoulos
Analysis of nearly 400,000 imaging studies highlights efficiency gains, ongoing integration barriers and radiologists’ perspectives on AI in clinical workflows.
An extensive analysis of AI deployment across 20 Swiss imaging centers revealed widespread adoption, measurable efficiency gains and persistent integration challenges, according to research presented by Sergey Morozov, MD, PhD, MPH, an independent AI consultant and the head of research and development at the 3R Swiss Imaging Network in Sion, Switzerland.
“This is one of the few works that analyze the large-scale practical implementation of AI in clinical radiology,” Dr. Morozov said. “It shows how combining objective data and personal experiences links technology, clinical practice and human behavior to improve both the quality of care and operational performance.”
Dr. Morozov and colleagues evaluated the four-year experience of nearly 400,000 AI-processed imaging studies collected between January 2021 and June 2025, spanning musculoskeletal, chest, breast and brain imaging. They combined quantitative data with a highly engaged radiologist survey to assess workflow impact, user sentiment and practical barriers to implementation.
“Across the 20-center network, AI tools have been used by 91% of radiologists, with musculoskeletal imaging representing the majority of use,” Dr. Morozov said. “Trauma radiography accounted for 37% of AI activity, while adoption in breast, chest and brain imaging reached 76% among specialists.”
AI applications yielded turnaround time reductions from 16% for knee MRI to 33% for trauma radiography.“This is one of the few works that analyze the large-scale practical implementation of AI in clinical radiology. It shows how combining objective data and personal experiences links technology, clinical practice and human behavior to improve both the quality of care and operational performance.”
Sergey Morozov, MD, PhD, MPH
Radiologists in the survey most often valued AI for its perceived protection against diagnostic errors and for time savings. However, many respondents called for faster AI computation and better report integration, with only one in five AI results available before reporting.
The Net Promoter Score, a measure of user satisfaction and loyalty, varied widely, from strongly positive for bone age (+86.7) and trauma radiography (+65.2) to negative for spine MRI (-55.3). This highlighted an uneven maturity of current AI tools.
Dr. Morozov said several findings reinforce long-held assumptions about effective AI implementation. He emphasized that much like a train, AI in radiology must be safe, precise, on time and convenient or risk being left behind.
“The timing of AI delivery is paramount, allowing for human-AI interaction exactly when demanded by the specific clinical workflow. If the best AI is not on time, it is worthless,” he explained.
Additionally, engaging radiologists in feedback and acknowledging their input proved to be a key factor in deep adoption.
The study underscored the need for interoperability and education. Dr. Morozov advocated for benchmarking initiatives and integration frameworks to build confidence and consistency in AI-supported radiology. “AI-native PACS and RIS systems are needed,” he said. “Formal AI education for medical school faculties is also needed to train residents based on the latest technical developments, rather than following hype.”
Dr. Morozov and his team plan to launch prospective studies on high-performing AI use cases, assess economic impact and expand into generative AI applications for report generation and opportunistic screening.
“Under the vision of leadership, including CEO Simon Pericou, COO Cyril Thouly and CMIO Benoit Rizk, 3R strives to make AI an essential component of quality and efficiency,” he said. “We want to make radiologists’ work more convenient, help experts help more patients, and grow the value of the Swiss Imaging Network.”
Access the presentation, “Large-Scale AI Implementation in Radiology: Technical, Operational and Behavioral Adoption Patterns Across a 20-Center Swiss Network,” (R3-STCE1-2) 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.
