ACR AI on the Global Stage

In February 2025, ACR leaders were invited to share the work of the ACR Data Science Institute^® (DSI) on artificial intelligence (AI) and the future of AI in healthcare at the Royal College of Radiologists’ (RCR) Global AI Conference.

This inaugural meeting gave ACR CEO Dana H. Smetherman, MD, MPH, MBA, FACR; Vice Chair of the ACR BOC and Chair of the ACR Commission on Informatics Christoph Wald, MD, PhD, MBA, FACR; and Chair of the ACR Commission on Quality and Safety and AI Advisory Group Member David B. Larson, MD, MBA, FACR, an opportunity to both share the ACR’s initiatives to ensure the safe and effective use of AI in US radiology practices and learn more about the state of AI around the world.

The Bulletin sat down with the three leaders to discuss the outcomes and importance of the meeting.

Could you share how ACR’s invitation to present at the RCR Global AI Conference came about and what it means for ACR on an international stage?

Smetherman: In July 2024, my first month as ACR CEO, the leadership of the RCR reached out about the opportunity to partner with their organization for the Global AI Conference. Our organizations have historically had a very strong relationship, and I was delighted that ACR received this invitation. The RCR leadership wanted this to be a truly global conference, and radiology societies from several other countries (including Germany, France, the Netherlands, Sweden, Italy, Canada, and Korea) had already come on board. We set up a meeting to learn about the conference and decided to participate shortly thereafter.

Dr. Wald’s presentation at the conference, the “AI Ecosystem Under the Data Science Institute (DSI) of the ACR,” highlighted the growing importance of data science in radiology. How does the DSI support radiologists in harnessing AI and data science, and what future advancements are you most excited about in this space?

Wald: We are focused on the safe and effective implementation of AI, helping practices add the technology and manage it.

ACR offers three main programs. We provide overarching support via the recognition program, ACR Recognized Centers for Healthcare AI (ARCH-AI). We have created best practices to follow and key building blocks of a responsive AI program that departments should be implementing. If a practice is doing all of this, they can be recognized for it.

Then we have a navigable online resource for imaging AI solutions. This is continuously curated by ACR staff to help members find what is available for commercial licensing in the U.S. market.

Assess-AI is ACR’s AI registry. It’s the first national registry for AI and is scaling quickly. A site can transmit data about an AI transaction into the ACR registry and receive analytical information about the performance of whatever AI tool is being used. This also allows radiologists to see how that local performance compares to other sites using the same or different technology for the same use case.

I’m most excited for tools that use generative AI and large language models. We’re looking at the capability of that technology, as those models are much more powerful. In fact, ACR has partnered with several key universities in the ACR’s Healthcare AI Challenge, our first attempt to stage generative AI through common radiology tasks and see how well that technology functions in that way.

Smetherman: Many radiologists are not sure where to begin when they are considering incorporating AI. Our recognition program, ARCH-AI, has established guidelines and a framework for radiologists to make informed decisions about AI algorithm selection and implementation in their practices. Because AI is evolving so quickly, a Learning Network has been developed among the facilities in the ARCH-AI program to facilitate knowledge sharing.

Assess-AI, on the other hand, will enable radiology facilities to not only track the function of AI algorithms over time, but eventually to compare local performance against national benchmarks and detect real-time when AI is working differently in their practice compared to other sites. The Healthcare AI Challenge provides a way for ACR members to interact with and rate the interpretive performance of foundation models on imaging studies in a “gamified” fashion. Any ACR member can gain experience and become familiar with foundation models in this low-stakes, artificial environment outside of a real-world setting.

From a quality and safety perspective, how does ACR ensure the safe and effective integration of AI in medical imaging, and why is that such a critical focus for ACR?

Larson: Because the technology is changing so rapidly, the first step to ensuring the quality and safety of AI in radiology is learning from the radiology practices that are leading in this space. This was the impetus for ARCH-AI. The program is made up of sites that have established basic infrastructure and processes to ensure quality and safety. ACR is convening the ARCH-AI Learning Community to facilitate a discussion among these sites so that they can learn from each other and we can learn from them. This includes topics such as AI governance and management, evaluation and selection, performance monitoring, education and training and so forth. There is currently wide variation in practice in how AI is managed, which is an advantage at this stage since it represents innovation occurring across the country. We intend to harvest those ideas.

Dr. Larson’s presentation on the “Pathway to Radiology Practice Accreditation for the Use of AI” outlined the importance of AI-specific accreditation. Can you walk us through how the ARCH-AI program will help radiology practices implement and monitor AI safely, and how it aligns with ACR’s long-term vision for accreditation by 2027?

Larson: At ACR, we believe that the safety and quality of AI will be managed primarily by the local practice, in conjunction with the local health system, similarly to imaging modalities and clinical programs. The role of ACR is to provide resources for local practices to establish their program, facilitate learning among sites and provide opportunities for quality improvement. Eventually, practices engaging in this effort will be able to apply for ACR accreditation to receive external validation of the quality and safety of their program in AI, as for other clinical programs, which both provides the opportunity for local practices to receive feedback and demonstrates the level of quality to external entities. The ARCH-AI program serves as a precursor to the ACR AI Accreditation Program as we gain the experience needed to define the infrastructure, programs and processes needed to run a high-quality clinical AI program in radiology. When local sites join ARCH-AI, they become part of the ARCH-AI Learning Community, where they can connect with and learn from other radiology practices that are working through similar challenges in figuring out the best way to evaluate, deploy and monitor AI in their clinical environments.

What were some of the key takeaways from the RCR conference on AI in medical imaging, and how is ACR staying ahead of global trends?

Wald: Our regulatory framework is more conservative in the U.S. than in other regions. So compared to other countries, the availability of products is a bit more limited. That’s the key message: AI is a global phenomenon, and what AI tools can be used in each region of the world and country depends on the applicable regulatory framework. Observing use of AI in other countries provides us with insight on which types of AI technology might be worth considering for later adoption in the U.S.

As leaders, when we’re going to a global conference, we’re there with our eyes wide open to get a good preview of what’s already being used in areas where the regulatory framework is more permissible. We’re learning how it’s working, whether it’s helping or harming patients and what practices are seeing. This can inform our own future and advocacy for or against certain approaches.

AI has the potential to significantly improve healthcare delivery. What do you see as the greatest opportunities and challenges when it comes to implementing AI in radiology across different healthcare systems?

Larson: Interestingly, the implementation of AI at scale will likely force health systems to develop discipline in the nuts and bolts of quality that are just assumed in modern businesses in other industries. For example, implementing AI across different healthcare systems requires developing a common naming convention for MR series to which all MR exams can be mapped for the AI applications. As another example, since AI depends on sufficient image quality, we will begin to see more quantitative methods for measuring and more rigorous process controls for ensuring image quality. As yet another example, the performance of diagnostic AI applications is closely tied to the positivity rate of the studies — the lower the positivity rate, the more false-positives returned by the AI. This will likely spur health systems to more closely monitor their positivity rate, which may help identify which studies are not indicated and may incentivize health systems to improve ordering appropriateness. As these AI implementation challenges are identified and overcome, it will increase standardization and interoperability, which will both facilitate AI implementation and identify new challenges, which will spur innovation that will further facilitate AI implementation, and so forth, in a virtuous cycle.

In what ways does ACR plan to continue its leadership in AI and collaborate with global organizations to ensure the responsible and effective use of AI in imaging worldwide?

Smetherman: In addition to the RCR Global AI Conference, several of our ACR leaders lectured and met with leaders from other international radiology societies at the 2025 European Congress of Radiology in Vienna. This provided another important opportunity to not only learn what other countries are doing but also showcase the important work of the College in AI monitoring. Like our experience at the Global AI Conference, there was tremendous interest in ACR’s AI initiatives. We plan to continue to work collaboratively with other radiology societies and share our progress and lessons learned in radiology AI performance monitoring. These efforts will be critical as AI adoption grows, not only in the U.S. but throughout the world.

There is still much work to be done to facilitate the safe, effective, and responsible use of AI in radiology — some of it at the local practice level. Innovation is moving at an incredibly fast pace in the AI space, and communication and collaboration across the field are crucial to keep up with these advances. ACR will continue to facilitate this type of learning and knowledge sharing across radiology practices and with other professional radiology societies to ensure the quality and safety of AI implementation keep pace with the rapid development of the technology. Fortunately, there is no shortage of volunteers and healthcare leaders stepping forward to help make this happen.

Interview by Raina Keefer, contributing writer, ACR Press