Radiology educators partner with RadClerk to study AI/LLMs in medical education

Berkeley, CA—Radiology educators at the world's leading medical institutions have launched a research project in collaboration with RadClerk to understand the potential of AI based on large language models (LLM) to scale radiology education for medical students. The new collaboration brings together radiology educators at Harvard/MGH, Yale, University of California San Francisco, and UChicago Medicine, with expertise that spans medical pedagogy, clinical practice, and administrative leadership.

The wide release of advanced large language models (LLM) in early 2024 has already made huge waves across many fields. But the implications for conversant AI agents in medicine are only beginning to be uncovered. And the potential for LLMs to transform medical education is almost entirely unexplored.

RadClerk's pioneering use of this technology promises to scale the expertise of radiology educators by giving every student with a browser the chance to experience a full-featured radiology clerkship. "We could not have predicted the capabilities of our virtual radiology clerkship just 12 months ago," said clerkship director Ram Srinivasan MD PhD, "the pace of innovation has been staggering."

Using the Voxl viewer in any standard browser, students navigate radiology studies with the same capabilities as professional radiology workstations. The built-in reporting templates and AI agent allows students to begin refining their ability to interpret imaging with the personalized feedback that is normally reserved for resident physicians in radiology. Early indications suggest this personalized approach may drive radiology skill acquisition with unprecedented efficiency, which could be relevant to both medical students and radiologists-in-training.

Over the course of this collaboration, the investigators plan to explore a wide range of fundamental questions related to the use of LLMs in medical education, related to safety, efficacy, and impact. The most immediate pilot will focus on student perceptions of AI agents as teachers. While RadClerk is already showing substantial capability in ongoing clerkships, quantifying attitudes will allow the researchers to determine potential for application more broadly.

— RadClerk Staff