At lunchtime on November 30, a flood of people crowded into Harkness Memorial Auditorium: Over 100 people had registered for Professor Andrew Taylor’s talk on "Bringing AI to the Bedside." Sponsored by the Section of Biomedical Informatics and Data Science, the event was co-hosted by two graduate student organizations: the Yale Journal of Biology and Medicine, and the new AI in Medicine Student Interest Group. It was the groups’ most highly attended event of the semester, and members of the crowd ranged from tenured professors to fresh-faced first-years who filed into the auditorium to learn about data architectures, scalable infrastructures, and best practices for reliable AI solutions.
Founded this summer, the AI in Medicine Student Interest Group at Yale School of Medicine arose from a climate of excitement around artificial intelligence and the role new technologies might play in improving medical research and care. Nisarg Shah, a medical student at Yale and the founder of the group, says he hoped to increase medical students’ exposure to the role of AI in healthcare. “Seeing the rapid expansion of the technology and its numerous practical applications in medicine, I believe that understanding artificial intelligence, even at a basic level, may become a necessary competency for the 21st century physician,” Shah says. Noticing a lack of structured support in this area, Shah decided that a specialized group was needed to expand AI’s footprint at Yale School of Medicine. Since its creation, the group has launched a monthly seminar series, with topics including privacy-preserving techniques and guests including Yale School of Medicine’s new Deputy Dean for Biomedical Informatics, Lucila Ohno-Machado. They are also working on a research project directory and other initiatives.
Philip Adejumo, MD-PhD student at Yale School of Medicine, is one of the group’s leaders. For him, one of the strengths of the group is its ability to create connections between faculty and interested students. “We’re trying to find ways to help students find labs and projects they can get involved in,” Adejumo says. The group also supports students who want to enter industry and biotechnology, acting as a network for them to find internships.
“We’re establishing our presence and getting people interested,” Joanna Fuyao Chen adds. A familiar face to anyone who attends the AI in Medicine monthly seminar series, Chen is an MD-PhD student and the point-person for coordinating seminar logistics with staff from the Section of Biomedical Informatics and Data Science. She attributes the success of the seminars to administrative support from the Section as well as David Chartash, a lecturer who has mentored students in the group. Clearly, the group’s success is a team effort.
Though they have been hard at work organizing projects and events, getting involved with the AI in Medicine group has also deepened Adejumo’s and Chen’s personal interests in artificial intelligence and healthcare. Now in the fifth year of her program, Chen has chosen to focus her research on how artificial intelligence, deep learning and image-processing can be used to diagnose prostate cancer. “Currently, the gold standard [for diagnosis] is needle biopsy, which is obviously not very pleasant!” Chen laughs. “And, you only get a little amount of tissue, so you’re missing out on the heterogeneity of the entire prostate.” Chen says that Multiparametric Magnetic Resonance Imaging (MRI) techniques, which are already used for pre-biopsy planning, could be used to predict whether or not a patient has indolent or aggressive prostate cancer. “Why not try to realize the full potential of this imaging procedure? With preliminary results, just using metadata and machine learning algorithms, we probably would be able to avoid half the current biopsies being done.”
Like Chen, Adejumo is excited by the ways that AI can be used to improve the quality of care patients receive. “Heart failure is a chronic disease that affects over 8 million Americans every year,” he says. The guidelines for treating heart failure involve the implementation of four different drugs, says Adejumo, but this standard isn’t applied as widely as it should be. As a result, he has become interested in using natural language processing (NLP) and machine learning (ML) to identify why hospitals and health systems are not implementing these drugs as efficiently as possible, and to find out why patients might not be using these drugs. “This is all in the realm of quality of care process measures,” Adejumo adds, “Meaning that we’re trying to improve the process of care that patients are seeing at the hospital, and we’re trying to use deep learning and machine learning to improve the capture and recording of these process measures in cardiovascular care.”
The more Chen and Adejumo investigate, the more they learn about how challenging it can be to implement these technologies in an effective way. “In the image processing community, the matrix they use to evaluate their results is not very intuitive to the clinical side,” Chen says. “It’s important to benchmark these models and tools in clinically meaningful ways, so that AI can expand its impact on the broader community.”
“There are challenges that people should be aware of, including privacy, disparities, hallucinations in Large Language Models, and other problems I wasn’t exposed to before,” Adejumo warns. “AI can falter and fail patients…[it’s] only as good as the humans that train it, and the data it gets. In medicine, that tends to be incredibly biased.” Adejumo has been spending time learning about the ways that AI can perpetuate disparities and discrimination—but also how it can be used to reduce inequities in health.
These challenges and opportunities reflect the wide range of topics that medical AI research can explore. Though the AI in Medicine Student Group has only been around for one semester, Chen says momentum is rapidly growing. “We are receiving a lot of interest from people asking, ‘How can I get connected?’” she says. Like her, Shah shares this excitement: “I’m looking forward to seeing how the community can create projects and initiatives that get students more involved.”