Current and Emerging Concepts with LLMs in Radiology: An Interview with Rajesh Bhayana, MD

From the initial requisition for medical imaging to the radiology report, Rajesh Bhayana, M.D., said there is an opportunity at each stage of the process for a large language model (LLM) to “improve or augment” what clinicians do.

In an interview with Diagnostic Imaging at the RSNA conference, Dr. Bhayana talked about recent research demonstrating the use of LLMs to offer reliable access to clinical history information, a time-consuming challenge that has daunted radiology since the 1960s.

“For instance, in oncology patients, we need to know the primary tumor, other tumors, treatment history, potentially prior imaging findings, a few other things, maybe tumor markers. So we extracted these things and when (the LLM) has a very narrow task that it’s doing and it's optimized the correct way, (the LLM) can facilitate that process,” maintained Dr. Bhayana, an assistant professor radiology and radiologist technology lead in the Joint Department of Imaging at the University of Toronto.

(Editor’s note: For additional coverage of the RSNA conference, click here.)

Dr. Bhayana has also been involved with the development of the LLM models Radiology Knowledge Research, which provides immediate answers to radiology-specific questions, and AI Attending Voice Mode (Navigating Radiology), which is geared toward radiology training along with the capacity to adapt to the language of the user.

“ … This is making learning interactive by actually scrolling through cases, but even more interactive by making you commit to a diagnosis, giving you pointed feedback and really simulating like you're working with an attending, but without the risk or fear of not looking smart in front of that attending,” explained Dr. Bhayana, the founder of Navigating Radiology. “It kind of lets you do that from the comfort of your own home, and that's what we think the future of education is: talking to an AI as you're scrolling through training images.”

(Editor’s note: For related content, see “Can ChatGPT Pass a Radiology Boad Exam?,” “New Study Examines Short-Term Consistency of Large Language Models in Radiology” and “Emerging Directions with Advances in Enterprise Imaging in Radiology.”)

For more insights from Dr. Bhayana, watch the video below.