DiagLink: A Dual-User Diagnostic Assistance System by Synergizing Experts with LLMs and Knowledge Graphs

TL;DR

This work tackles the mismatch between healthcare resource limitations and diagnostic needs by proposing DiagLink, a dual-user diagnostic assistant that unites LLMs, medical knowledge graphs, and expert oversight. It features a role-adaptive, emotionally supportive patient interface, a physician-focused KG-augmented reasoning workflow, and an expert-in-the-loop process that continuously evolves the KG. Through a controlled user study with simulated patients and practicing physicians, DiagLink demonstrates enhanced patient satisfaction, reduced physician workload, faster and more accurate diagnostic performance, and strong safety signals with interpretable evidence. The approach has practical implications for improving diagnostic experiences and equity in care, and offers a template for future human-centered AI systems that merge generative language, structured knowledge, and expert governance in high-stakes domains.

Abstract

The global shortage and uneven distribution of medical expertise continue to hinder equitable access to accurate diagnostic care. While existing intelligent diagnostic system have shown promise, most struggle with dual-user interaction, and dynamic knowledge integration -- limiting their real-world applicability. In this study, we present DiagLink, a dual-user diagnostic assistance system that synergizes large language models (LLMs), knowledge graphs (KGs), and medical experts to support both patients and physicians. DiagLink uses guided dialogues to elicit patient histories, leverages LLMs and KGs for collaborative reasoning, and incorporates physician oversight for continuous knowledge validation and evolution. The system provides a role-adaptive interface, dynamically visualized history, and unified multi-source evidence to improve both trust and usability. We evaluate DiagLink through user study, use cases and expert interviews, demonstrating its effectiveness in improving user satisfaction and diagnostic efficiency, while offering insights for the design of future AI-assisted diagnostic systems.

Figures (9)

• Figure 1: DiagLink synergizes LLMs, KGs, and clinical experts to support a dual-user (patients and physicians) diagnostic workflow. It consists of three core modules: (1) a Dialogue-Driven Medical History Collection Module that interactively gathers structured patient data; (2) a Collaborative Diagnosis Module where LLMs and KGs perform parallel reasoning and retrieval-augmented ranking to generate candidate diagnoses, all under a Physician-in-the-Loop mechanism for validation and continuous KG evolution; and (3) a Dual-User Interface Module that provides an emotionally supportive guided dialogue interface for patients and an evidence-rich, KG-visualization dashboard for physicians. Arrows indicate the primary flow of data and reasoning, also represent expert-driven feedback and knowledge updates, forming a closed-loop collaborative framework.
• Figure 2: Physician interface, consisting of three components: patient information (A), medical knowledge (B), and diagnostic reasoning (C). The interface uses a flat layout with highlighted linked entities to integrate heterogeneous data, allowing physicians to quickly understand patient status and diagnostic context. Direct communication with patients is also supported when needed.
• Figure 3: Patient Interface.DiagLink provides a simple and intuitive conversational interface for collecting medical history while dynamically displaying the collected information on the right.
• Figure 4: Expert Interface. Expert can iteratively update, validate the DiagLink's knowledge. Newly added or modified components are highlighted to emphasize recent changes, while previously entities remain visible in a subdued form. This design supports focused editing while preserving awareness of the overall structure.
• Figure 5: Entity Exploration. Users can explore knowledge about entities in three forms: definition nodes, drug nodes, and shortest-path nodes connecting diseases and symptoms. This design balances information richness with cognitive load, allowing users to access relevant details without being overwhelmed.