Abstract No. 92: Dr. Boardwell: Development of an AI Interventional Radiology Expert for Real-Time Conversational Oral Boards Simulation and Case-Based Resident Training
An AI-powered conversational tool named “Dr. Boardwell” may be the future of IR oral board preparation.
Dr. Boardwell, developed by Kevin Seals, MD, and colleagues at Cedars-Sinai Medical Center and UCLA, is designed to simulate a board-certified IR whom residents can verbally converse with.
The first use case of the technology, as documented in Dr. Boardwell: Development of an AI Interventional Radiology Expert for Real-Time Conversational Oral Boards Simulation and Case-Based Resident Training, an SIR 2025 Featured Abstract, is of an oral boards simulator. The platform aims to provide trainees with a more effective way to prepare for their oral boards by closely replicating the actual exam experience through real-time verbal interaction.
The motivation behind this project stemmed from two key ideas, according to Dr. Seals. "First, IRs-in-training (and soon all radiology trainees) need to take oral boards, and there is no great way to prepare for them,” said Dr. Seals. “You can read and learn tricks to answer a multiple-choice question with limited knowledge, but you need much deeper knowledge to actually converse.”
Because the latest wave of generative AI is enabling physicians to push technological boundaries, Dr. Seals thought it could be a solution.
“We can now create an AI examiner with an excellent IR knowledge base that can verbally talk with trainees in real time,” he said. “They can now get the best possible oral boards preparation experience at their leisure, both developing oral exam skills and learning IR in a fun, interactive way.”
The Dr. Boardwell system integrates 10 de-identified cases covering common IR topics into an AI examiner powered by the OpenAI GPT-4 model via an application programming interface (API). Real-time verbal interaction is facilitated using Deepgram’s text-to-speech and speech-to-text APIs. The AI then analyzes residents' responses and generates detailed feedback summaries.
The initial pilot study included a cohort of 14 IR residents, and the posttest survey revealed high average scores for ease of use (4.8), fidelity (4.1), accuracy (4.3), flow (4.1), flexibility (4.6), educational value (4.9) and overall rating (4.6). According to respondents, the AI effectively presented cases, asked targeted questions and facilitated dynamic conversational interactions with relevant follow-up questions and redirection.
According to Dr. Seals, this research—as well as the high reviews provided by the trainees—provides the data needed to establish the validity of novel, AI-based learning techniques.
“This increases our confidence that these novel educational tools can have a big impact and fundamentally improve education in IR,” he said. “This results in smarter, happier trainees that go on to become better doctors and offer better patient care.”
Dr. Seals says Dr. Boardwell is only part of the growing role that AI will play in medical education.
"Soon it is likely that every student will have access to a personal AI tutor, basically democratizing the sorts of tools that were previously only available to wealthy families,” he said. “This sort of tutoring is proven to significantly improve achievement, and making it available to everyone is huge.”
Even though Dr. Seals and his team spent more than 1,000 hours developing the project, the technology is continually enhancing, especially as generative-AI techniques grow both more advanced and more cost-efficient.
The next step for Dr. Boardwell will be a larger study utilizing the more advanced software, which researchers plan to submit to an academic journal.
Beyond oral board preparation, Dr. Seals and his team are exploring other potential applications for this technology, such as practicing the consenting process.
“Some trainees may be asked to consent patients before they are great at it, and a tool like this can generate data for training programs verifying their resident can consent appropriately, credentialing them before they do it, and give the resident real practice and experience before they consent in the real world with real patients,” Dr. Seals said.
However, the applications for an AI-powered virtual examiner are endless, he said—as are the potential benefits to residents and, ultimately, patients.
Dr. Seals will present his findings at SIR 2025 on Monday, March 31, at 3 p.m. during the AI, Machine Learning, Simulation, and Imaging 1 session.
