GraphDx: A Cost-Aware Knowledge-Enhanced Multi-Agent Framework for Sequential Diagnosis

GraphDx：一种用于序贯诊断的成本感知知识增强多智能体框架

Sequential diagnosis requires balancing diagnostic accuracy against resource costs through iterative information gathering. Existing Large Language Model (LLM) approaches exhibit a critical knowledge-reasoning gap: despite encoding extensive medical knowledge, they struggle to reason systematically under cost constraints, often resorting to excessive testing.

序贯诊断需要在迭代信息收集的过程中，平衡诊断准确性与资源成本。现有的大型语言模型（LLM）方法存在一个关键的知识-推理鸿沟：尽管它们编码了丰富的医学知识，但在成本约束下进行系统性推理时往往表现不佳，经常导致过度检查。

We propose GraphDx, a knowledge-enhanced framework with two core innovations. First, we design an automated pipeline that leverages LLMs to construct Medical Diagnosis Knowledge Graphs (MDKGs) with quantized typicality, action-centric topology, and dual-objective attributes for both diagnostic relevance and cost-sensitivity.

我们提出了 GraphDx，这是一个具有两项核心创新的知识增强框架。首先，我们设计了一个自动化流水线，利用 LLM 构建医学诊断知识图谱（MDKG），该图谱具备量化的典型性、以行动为中心的拓扑结构，以及兼顾诊断相关性和成本敏感性的双目标属性。

Second, we introduce three collaborative agents (Perception, Reasoning, and Decision) where the Perception and Decision Agents handle language understanding and generation, while the Reasoning Agent performs deterministic evidence scoring and cost-aware planning on the MDKG.

其次，我们引入了三个协作智能体（感知、推理和决策）：其中感知智能体和决策智能体负责语言理解与生成，而推理智能体则在 MDKG 上执行确定性的证据评分和成本感知规划。

Experiments on MedQA and MIMIC-IV across three LLM backbones (DeepSeek-V3, Kimi-k2, Llama-3.3) show that GraphDx improves diagnostic success rates from 50—68% to 79—93% while reducing test costs by 20—54%, providing a robust, economical, and interpretable solution for automated clinical diagnosis.

在 MedQA 和 MIMIC-IV 数据集上，基于三种 LLM 主干模型（DeepSeek-V3、Kimi-k2、Llama-3.3）的实验表明，GraphDx 将诊断成功率从 50—68% 提升至 79—93%，同时将检查成本降低了 20—54%，为自动化临床诊断提供了一种稳健、经济且可解释的解决方案。