This event in the AI Forum includes the following presentation:
“Large Language Models and Medical Knowledge Grounding for Diagnosis Prediction”
By Yanjun Gao, PhD, Postdoc Research Associate, Critical Care Medicine (ICU) Data Science Lab, School of Medicine and Public Health, University of Wisconsin-Madison
Slides here: https://www.feinberg.northwestern.edu/sites/artificial-intelligence/docs/20231215aiforumyanjungao.pdf
View a summary of the presentation here: https://www.feinberg.northwestern.edu/sites/artificial-intelligence/docs/20231215aiforumsummary.pdf
Abstract:
The integration of Large Language Models (LLMs) into healthcare diagnostics represents a frontier in medical technology. However, the critical demand for minimizing diagnostic errors and ensuring patient safety necessitates a reliable and knowledgeable approach. This talk explores the innovative synergy between LLMs and Medical Knowledge Graphs (KGs), particularly focusing on grounding LLMs’ diagnostic processes in robust, structured knowledge. The first part of the talk will introduce a novel graph prompting method that utilizes a knowledge graph derived from the National Library of Medicine’s UMLS, an extensive source for medical concepts and relations. In the latter part, the talk will present the first comprehensive human evaluation for LLM for diagnostic safety, assessing the quality of LLM’s predicted diagnosis as well as its reasoning behind the diagnostic process. Our discussion will cover the impact of incorporating external knowledge sources into LLMs to enable explainable diagnostic pathways, highlighting both the hurdles and the prospects for AI-augmented diagnostic decision support systems.
Add comment