Curtis P. Langlotz, MD, PhD
The growing role of artificial intelligence (AI) in radiology has been increasingly showcased during Radiological Society of North America Scientific Sessions and Annual Meetings, and with programs and certifications offered year-round. After RSNA 2023, held Nov. 26-29 in Chicago, I interviewed 2023-2024 RSNA President, Curtis P. Langlotz, MD, PhD, of Stanford University, a renowned imaging informatics leader. He offered insights into how radiologists can leverage AI to support their work and patient outcomes, and addressed key issues. Portions of that discussion follow. You can view the full video interview here.
Q: Can you speak about RSNA 2023 and what you feel is most important to know now?
LANGLOTZ: I think you could say the RSNA annual meeting is back. We had 40,000 attendees, 700 companies, many new vendors, in over 400,000 square feet of tradeshow space. The atmosphere and the buzz was very much like 2019. The meeting was just as it had been, and it was just very, very exciting to be there, including an emphasis on artificial intelligence (AI), which has been featured at recent meetings and continues to be. We had a very nice AI Showcase with about 100 companies. I think that’s going to continue to be a very active area, both for the RSNA and for the field generally.
Q: During the Opening Plenary, your quote “AI won’t replace radiologists, but radiologists who use AI will replace those who don’t” was referenced. What insights do you offer early career radiologists looking to responsibly and effectively work with AI?
LANGLOTZ: That quote was a tweet of mine from back in 2017 when there was a lot of concern about whether AI will replace radiologists, and I think we now know that that those fears are really unfounded. These systems are likely to be a partner and a help for us. It’s really interesting if you think about recent technologies, even electronic health records (EHR), which has put kind of a burden on some of our clinical colleagues, and even PACS, which is a great new technology, but actually tends to isolate us … radiology used to be the place where everyone needed to go to look at their images and talk to the radiologist. Now the images can be viewed from anywhere and we don’t get as much of that contact.
Part of that is due to technology. The technology in the past has served to, in some ways, reduce our human connections. However, I think that with the latest wave, some of these newer AI technologies really have the opportunity to increase our connections to one another. For example, some of these large language models like ChatGPT could be used to explain our test results. Our reports, for good reason, are written as a communication to another physician, so there is a lot of complex terminology there. With this new legislation requiring that patients get digital access to their medical records, patients are seeing these radiology reports, but they don’t always understand what they’re seeing. So now we can use these new technologies to maybe rephrase those terms in a in a manner that they can understand, and that can create potential for better connection between patients and their radiologists.
Also, some of the AI methods can actually make us more efficient, giving us time for things that we would prefer to do. Some of these ‘needle in a haystack’ problems — finding nodules, finding tiny abnormalities, that’s an area where AI can really help us. There are technologies that I think will be coming very soon, that can draft a report for us, much as a human radiology resident would do. This can free up time for us to relate to both our radiology colleagues and our clinical colleagues. I would summarize by saying in the past, we’ve tended to think about high tech and high touch as two ends of a spectrum. But I think that increasingly, at least with the technologies that we’re seeing now, some of these new AI methods will actually help to bring us closer together rather than split us apart.
We’re at a point in the evolution of the radiology AI market, where vendors and those who develop AI algorithms are thinking more about what the problems are that radiologists really need to have solved. Where do we need help? I think in the past, sometimes AI algorithms were developed in part just because there was data available to train the algorithm, or maybe because it could make a splash, or could easily make it through regulatory approval.
So I think that’s a healthy development, and I think we’ll see more of that, with more solutions to which radiologists really gravitate.
Q: For what are you most excited? About what are you most concerned?
LANGLOTZ: For radiology generally, I’m obviously very excited about AI and what it could do for our profession. I think that image-guided intervention is a huge area of growth, and an area where we can reduce morbidity to patients by making interventions less invasive. And then there is theranostics, where we can target individual tissues or cells within the within the body and, in fact, use that not just for diagnostic imaging, but for patient treatment. I think those are huge opportunities.
And then to challenges. I think the workforce is a major challenge. We have declining reimbursements, increasing volumes and not enough radiologists. I think we’re going to need a multifactorial approach to how we face that. That also can lead to burnout among radiologists. That’s where I think some of the newer technologies might actually help to reduce burnout with all the other methods that we apply to that.
Q: What are the top priorities with RSNA in the coming year?
LANGLOTZ: There is so much going on with RSNA. With the annual meeting, we obviously have several priorities with AI. We have an AI certificate program, which should be really useful for anyone who wants to learn more when asking: “How do I decide when to implement AI in my practice? How do I work optimally with AI algorithms that are already implemented in my practice?” There are basic and advanced certificates, and those will be a focus.
RSNA has also started a program called Global Learning Centers. We have learning centers in Tanzania, Ecuador, Indonesia and a new one in Armenia.
Our journals will continue to be a focus, and we have really strong impact factors from all of our journals. We have a new open journal called Radiology Advances starting this year. We have a new educational platform called Ed Central. This uses what’s called collaborative filtering. This is much like what Netflix uses to recommend movies based on what your interests are, what you’ve already watched. With this, RSNA members can go there and express a little bit about their preferences for learning. As one continues to use the system and take advantage of the learning opportunities there, it will begin to recommend things that are useful to them in their ongoing learning.
We’ll continue with our research grant programs through the RSNA Research and Education Foundation. Last year, we gave $4.7 million in grants, including research grants, and that will continue as well. Also, we’ve started a Sustainability Task Force at RSNA which I think will be very important.
These are all key programs on which we’ll be focused, together with all of the wonderful human connections that we create with our meeting and throughout the year.
Q: Can you share with us the work on which you and your colleagues at Stanford are currently focused?
LANGLOTZ: In the lab, there are so many wonderful research opportunities these days. We’re focused on a couple of things.
One is what are called foundation models. The same methods that trained chat GPT — where you take a very large amount of text like the entire internet, and then you have the computer play a game with itself so it hides one word, and then tries to predict the missing word. We can do the same thing in medicine. For example, for medical imaging, we can obscure a part of the image and ask it to predict the missing piece. Or, we can feed it an image and a report and ask “Does this report match the image or is it a mismatch?”
Through doing that, the algorithm actually learns very well how to encode and process information both from medical images and from medical text. So I don’t think we’ve yet to have a kind of “ChatGPT moment.” It’s just amazing what happened when ChatGPT came out. We realized that just by predicting the next word over and over again, and learning from a large amount of text, it can do some amazing things. I think that we have yet to train on datasets that are as large as that, in medicine. So we’re really working on those kind of medical foundation models in the same way that they have been pursued outside of medicine.
The other thing that I think is very important and that we’ll be working on is multimodal. We see algorithms that process images, we see algorithms that process text. I think the real power is going to come from algorithms that process all of these data together — from genomics, wearables, EHR data, lab results, text notes, and of course images, both radiology, pathology and all the other imaging modalities … so, a focus on bringing all of that data together, analyzing it and coming up with the best answer for patients.
Then finally, there’s just so much work going on in determining how we responsibly implement artificial intelligence. We have a new initiative at Stanford, sponsored by our dean and the director of our university-wide Institute, which is called RAISE: Responsible AI for Safe and Equitable Health. It deals with all of those issues — ethics, fairness, accuracy, utility, all the things that we need to assess as we implement these to make sure that they’re safe and effective for our patients. So there is definitely a lot going on at Stanford.
SIDEBAR:
Curtis Langlotz, MD, PhD
Described by RSNA in a published news report as “a renowned imaging informatics leader and committed advocate for improved clinical communication,” Langlotz began his term as president of RSNA during the society’s recent Annual Meeting. Highlights of his career follow.
Current positions held at Stanford University:
• Director of the Center for Artificial Intelligence in Medicine and Imaging (AIMI Center)
• Professor of Radiology, Medicine, and Biomedical Data Science
• Associate Director of the Institute for Human-Centered Artificial Intelligence (HAI)
• Associate Chair for Information Systems, Medical Informatics Director for Stanford Health Care
RSNA Leadership:
• Began 2023-2024 term as President of the Radiological Society of North America Board of Directors during the organization’s 109th Scientific Session and Annual Meeting, held Nov. 26-29, 2023 in Chicago
• Served on the RSNA Board of Directors since 2016 as Liaison for Information Technology and the Annual Meeting, and most recently chaired the Board of Directors
• A longtime RSNA member, serving for many years on RSNA’s Radiology Informatics Committee, also serving the Society as an informatics advisor
• Led the development of numerous RSNA informatics initiatives, including the RadLex terminology standard, the LOINC-RadLex Playbook of standard exam codes and the RSNA imaging AI certificate program
• Served as a member of the RSNA Publications Council, the Research Development Committee, the Radiology Editor Search Committee and the Steering Committee for the RSNA Digital Roadmap
Medical Society and Organization Involvement:
• A founder and past president of the Radiology Alliance for Health Services Research (RAHSR)
• Former president of the Society for Imaging Informatics in Medicine (SIIM), and the College of SIIM Fellows
• Former board member of the Association of University Radiologists (AUR), the American Medical Informatics Association (AMIA) and the Society for Medical Decision Making (SMDM)
• Recipient of the Lee B. Lusted Research Prize from the Society of Medical Decision Making and the Career Achievement Award from the Radiology Alliance for Health Services Research
• Recipient of numerous scientific awards, along with his trainees, including seven best paper awards and five research career development grants
Education and Background:
• Received his medical degree, a master’s degree in artificial intelligence and a doctorate in medical information science from Stanford University, accepting his current position in 2014
• Completed an internship and radiology residency at the University of Pennsylvania, where he remained on the faculty for 20 years
• Founded several healthcare information technology companies, including Montage Healthcare Solutions, which was acquired by Nuance Communications in 2016
•Raised in St. Paul, Minnesota
Principal Investigator for several projects funded by the National Institutes of Health (NIH), including the Medical Imaging and Data Resource Center (MIDRC), an open-source database containing medical images from over one hundred thousand COVID-19 patients to help doctors better understand, diagnose, monitor and treat COVID-19.
Authored “The Radiology Report: A Guide to Thoughtful Communication for Radiologists and Other Medical Professionals;” co-edited “Cancer Informatics: Essential Technologies for Clinical Trials;” published over 150 scholarly articles.
Related content:
VIDEO: One on One with Curtis P. Langlotz, MD, PhD, RSNA President
April 25, 2024 
