SIR17: AI chatbot can answer questions about interventional procedures

The same technology that allows Google Translate to “help” high school students with their Spanish homework may soon put an interventional radiologist in your pocket, according to a March 8 session at the Society of Interventional Radiology’s 2017 Annual Scientific Meeting.

Researchers from the University of California-Los Angeles (UCLA) used natural language processing (NLP) to create a chatbot that can answer interventional radiology questions posed by referring physicians, allowing those physicians to provide timely and evidence-based answers to patients.  

“This research will benefit many groups within the hospital setting. Patient care team members get faster, more convenient access to evidence-based information; interventional radiologists spend less time on the phone and more time caring for their patients; and, most importantly, patients have better-informed providers able to deliver higher-quality care,” co-author Kevin Seals, MD, resident physician in radiology at UCLA and the programmer of the application, said in a statement.

At its core, NLP analyzes language for meaning using the hierarchical structure of language. Words make phrases, phrases make a sentence, and sentences convey ideas. According to John Rehling, an NLP expert at the Meltwater Group, what sets NLP apart is the ability to develop a knowledge base through machine learning.

Instead of hand-coding large sets of rules, developers can feed hundreds or thousands of examples—often real-world data—and the NLP application will learn rules through statistical inference.

Seals and his colleagues taught the chatbot about interventional radiology by feeding it more than 2,000 data points simulating common inquiries a radiologist might receive during a consultation. This resulted in a prototype that can respond to naturally-phrased questions with websites links, infographics, and even custom programs.

The small team of hospitalists, radiation oncologists, and interventional radiologists who are testing the chatbot say the prototype has promise.

“I believe this application will have phenomenal potential to change how physicians interact with each other to provide more efficient care,” John Hegde, MD, resident physician in radiation oncology at UCLA, said in a statement. “A key point for me is that I think it will eventually be the most seamless way to share medical information. Although it feels as easy as chatting with a friend via text message, it is a really powerful tool for quickly obtaining the data you need to make better-informed decisions.”

The researchers hope to next expand the chatbot beyond interventional radiology, hopefully fulfilling a similar consulting role for other specialties such as cardiology or neurology.