KERAP: A Knowledge-Enhanced Reasoning Approach for Accurate Zero-shot Diagnosis Prediction Using Multi-agent LLMs

Click to View Presentation

2025 Annual Symposium On Demand

Presentation Time: 08:00 AM - 08:15 AM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Knowledge Representation and Information Modeling, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

Medical diagnosis prediction plays a critical role in disease detection and personalized healthcare. While machine learning (ML) models have been widely adopted for this task, their reliance on supervised training limits their ability to generalize to unseen cases, particularly given the high cost of acquiring large, labeled datasets. Large language models (LLMs) have shown promise in leveraging language abilities and biomedical knowledge for diagnosis prediction. However, they often suffer from hallucinations, lack structured medical reasoning, and produce irrelevant or useless outputs. To address these challenges, we propose KERAP, a knowledge graph (KG)-enhanced reasoning approach that improves LLM-based diagnosis prediction through a multi-agent architecture. Our framework consists of a linkage agent for attribute mapping, a retrieval agent for structured knowledge extraction, and a prediction agent that iteratively refines diagnosis predictions. Experimental results demonstrate that KERAP enhances diagnostic reliability efficiently, offering a scalable and interpretable solution for zero-shot medical diagnosis prediction.

Speaker:
Yuzhang Xie, Doctoral Student
Emory University

Authors:
Yuzhang Xie, Doctoral Student - Emory University; Hejie Cui, Postdoc - Stanford University; Ziyang Zhang, Doctoral Student - Emory University; Jiaying Lu, PhD - Emory University School of Nursing's Center for Data Science; Kai Shu, PhD - Emory University; Fadi Nahab, MD - Emory University; Xiao Hu, PhD - Emory University; Carl Yang, PhD;

---

Model Quality in AI-based Bruise Detection: Rethinking IoU and Confidence Thresholds

Click to View Presentation

2025 Annual Symposium On Demand

Presentation Time: 08:15 AM - 08:30 AM

Abstract Keywords: Artificial Intelligence, Imaging Informatics, Fairness and elimination of bias, Deep Learning
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

Artificial Intelligence (AI)-based object detection models, like YOLO and Faster R-CNN, depend heavily on the Intersection over Union (IoU) and confidence thresholds to evaluate model performance. However, fixed thresholds can be biased and may have a disparate effect across subpopulations, even when traditional performance metrics suggest strong model performance. This paper examines how varying IoU and confidence thresholds affect standard evaluation metrics like precision, recall, F1-score, and mean Average Precision (mAP) along with their effect on five widely used fairness metrics: Demographic Parity, Equalized Odds, Equality of Opportunity, Accuracy Equality, and Disparate Impact. This study tested a dataset of bruise images under natural and alternative light sources and found that fairness and performance trade-offs can be mitigated by selecting intermediate threshold values rather than fixed thresholds. In addition, the results highlight the need for dynamical optimization of thresholds to achieve both, high model performance and fairness, in AI-driven bruise detection.

Speaker:
Dharmi Desai, MS
George Mason University

Authors:
Janusz Wojtusiak, PhD - George Mason University; Amin Nayebi Nodoushan, PhD - George Mason University; David Lattanzi, PhD - George Mason University; Katherine Scafide, PhD, RN - George Mason University; Mehrdad Ghyabi, PhD - George Mason University; Dharmi Desai, MS - George Mason University; Michał Markiewicz, PhD - Jagiellonian University;

---

Utilizing the Decedent Affairs Navigator to Ensure High Reliability Communication and Morgue Processes

Click to View Presentation

2025 Annual Symposium On Demand

Presentation Time: 08:30 AM - 08:40 AM

Abstract Keywords: Informatics Implementation, Change Management, Documentation Burden, Interoperability and Health Information Exchange, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Background: As Hackensack Meridian Health (HMH) experienced an influx of deceased patients during the COVID-19 pandemic three circumstances recurred: 1) the location of deceased patients were not being tracked properly, 2) the belongings and bodies of the deceased were being lost throughout the system, and 3) bodies were being released before autopsies were performed. HMH took on the U.S. Surgeon General’s challenge to reduce healthcare worker burnout by developing a morgue tracking process called Decedent Affairs.

Objectives: To standardize the postmortem checklists for the adult, pediatric, neonatal and fetal demise populations and consolidate centralized documentation in a Decedent Affairs Navigator. This would promote transparency in the morgue tracking process, reduce redundancy in documentation, improve overall burnout, and aid in HMH’s devotion to being a high reliability organization (HRO).

Method: A multidisciplinary team customized the Decedent Affairs navigator to HMH’s needs. Key features included Shared Workspaces, Reduction of Documentation Burden, and Organization of Decedent Tracking. The navigator went network-live on October 18th, 2023. Implementation approaches included the use of on-site workshops and digital tutorials.

Results: There was an 83 % rate decrease in the number of incidents of missing decedent items and/or bodies. Duplicative documentation reduced by 11.4%, saving documentation time by 31 seconds.

Conclusion: The Decedent Affair navigator offers a niched and centralized hub within Epic to document information related to death procedures that can simplify workflows and reduce frequencies of lost bodies and personal properties of deceased patients in healthcare organizations.

Speaker:
Ijeoma Okeke, MD, MHA
Jersey Shore University Medical Center

Authors:
Joy Mamuszka, MSN, RN-BC - Hackensack Meridian Health; Lauren Koniaris, MD - Hackensack Meridian Health;

---

A Reinforcement Learning (RL)-Motivated Simulation Framework for Evaluating Vancomycin Dosing Strategies

Click to View Presentation

2025 Annual Symposium On Demand

Presentation Time: 08:40 AM - 08:55 AM

Abstract Keywords: Deep Learning, Precision Medicine, Infectious Diseases and Epidemiology, Patient Safety, Artificial Intelligence, Quantitative Methods
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

Achieving and maintaining the therapeutic range in vancomycin treatment is important for optimal outcomes. While guidelines and best practices based on empirical studies exist, the theoretical best dosing strategies under various conditions remain illusive. We developed an RL-based simulation framework using a deep learning two-compartment pharmacokinetic model (PKRNN-2CM) and introduced the area under the time-concentration curve (AUC) reward score, which translates clinical guidelines into an RL reward. Ground truth time-concentration curves were generated from patient-specific data, and simulated curves were produced under different dosing strategies with optional noise perturbations to mimic real-world settings. Evaluation metrics included 24-hour AUC assessments and RMSE. Results indicated that while the low-dosing AUC target (low-doser) and the high-dosing AUC target (high-doser) performed comparably in noise-free conditions, the low-doser achieved slightly higher AUC reward scores under noisy conditions, whereas the high-doser exhibited greater stability. This framework opens new approaches for optimizing vancomycin dosing.

Speaker:
Degui Zhi, Ph.D.
The University of Texas Health Science Center at Houston (UTHealth) McWilliams School of Biomedical Informatics

Authors:
Bingyu Mao, MA - The University of Texas Health Science Center at Houston (UTHealth) School of Biomedical Informatics; Ziqian Xie, Ph.D. - The University of Texas Health Science Center at Houston, Houston, Texas, United States; Laila Rasmy, PhD, MSc, MBA, RPh. - UTHealth MSBMI; Masayuki Nigo, M.D. - Houston Methodist Hospital; Degui Zhi, Ph.D. - The University of Texas Health Science Center at Houston (UTHealth) McWilliams School of Biomedical Informatics;

---

H&E Referenced Multiplex Immunofluorescence Interpretation using Image Registration and Virtual H&E Generation

Click to View Presentation

2025 Annual Symposium On Demand

Presentation Time: 08:55 AM - 09:10 AM

Abstract Keywords: Imaging Informatics, Computational Biology, Machine Learning, Evaluation
Working Group: Biomedical Imaging Informatics Working Group
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF

Multiplex immunofluorescence (MxIF) images provide an effective way to investigate cell interplays within cancer tissue by revealing biomarkers in different cell types, but the integrity of the biomarker signals can be compromised due to complex cyclic staining protocols. We proposed to use clinical standard H&E images to validate marker signals by aligning the two image modalities. Virtual H&E generation models were also developed and evaluated for cases without real H&E.

Speaker:
Jun Jiang, Ph.D.
University of Texas Health Science Center at Houston

Authors:
Raymond Moore, M.S. - Mayo Clinic; Brenna Novotny, M.S. - Mayo Clinic; Ruifeng Guo, M.D. Ph.D. - Mayo Clinic; Zachary Fogarty, M.S. - Mayo Clinic; Yuanhang Liu, Ph.D. - Mayo Clinic; Ellen Goode, Ph.D. - Mayo Clinic; Stacey Winham, Ph.D. - Mayo Clinic; Hongfang Liu, PhD - University of Texas Health Science Center at Houston; Svetomir Markovic, M.D. Ph.D. - Mayo Clinic; Chen Wang, PhD - Mayo Clinic;

---