🤖 AI Frontiers Seminar: Hosted by NLP/LLM Interest Group (Virtual Webinar)

From Lab Values to Clinical Reasoning: Laboratory Test Interpretation in the Era of Large Language Models by Balu Bhasuran, PhD, Research Faculty in the School of Information, College of Communication and Information at Florida State University.

Abstract: Large language models (LLMs) are increasingly being used to interpret laboratory test results. However, laboratory test interpretation remains a complex clinical reasoning task that requires more than reference-range knowledge alone. Accurate interpretation must integrate patient context, longitudinal trends, clinical reasoning, and an understanding of uncertainty. In this talk, Bhasuran will present a research program examining how LLMs interpret laboratory tests across patient-facing, diagnostic, and reasoning-centered settings. He will begin with studies evaluating patient-facing LLM-generated explanations of laboratory results, followed by the development of benchmark datasets such as LabQAR and retrieval-augmented generation systems designed to improve contextual accuracy. He will then discuss how laboratory results influence LLM-generated differential diagnosis, the role of causal reasoning in laboratory test interpretation, and recent work using EHR-derived patient profiles to generate personalized Question Prompt Lists. Together, these studies motivate a future vision of safe, retrieval-grounded, and clinician-supervised AI systems that empower patients and support clinical decision-making.

Balu Bhasuran, PhD, is a Research Faculty I in the School of Information, College of Communication and Information, at Florida State University. He previously served as a Visiting Assistant Professor and Postdoctoral Fellow in the eHealth Lab at Florida State University under the mentorship of Prof. Zhe He. Before joining FSU, he was a Postdoctoral Fellow at the Bakar Computational Health Sciences Institute at the University of California, San Francisco.

Dr. Bhasuran's research focuses on clinical natural language processing, machine learning, generative AI, information extraction, and clinical data science. Over the past decade, he has developed classification and prediction models using biomedical and clinical data for a range of applications, including early diagnosis of rare diseases, disease severity classification, adverse drug event detection, organ transplant rejection prediction, and medication adherence modeling. His broader research interests include causal inference, literature-based discovery, knowledge graphs, network visualization, fairness, and explainability in machine learning.