跳转至

EviDx：基于脚手架大模型智能体的证据感知主动诊断框架

文章背景与核心概要

临床诊断本质上是一个主动且迭代的过程，临床医生在此过程中收集证据、更新竞争假设，并在信息充分时做出明确诊断。然而，传统的医学大语言模型系统通常将诊断视为静态的单轮“病例-答案”任务，缺乏动态获取证据的能力。尽管具备自主性的智能体（Agentic LLM）可以通过工具使用提供动态交互，但它们往往缺乏对患者证据的精确运行时控制与脚手架支撑。

为了弥补这一空白，本文推出了 EviDx——一个全新的证据感知主动诊断框架。该框架有机结合了三大核心组件：1. 通过 \(\mathcal{E}\)-Synthesis 从原始临床病例中构建出患者专属的诊断环境；2. 组织了角色专用智能体、证据工具以及演进证据状态的临床诊断脚手架；3. 通过持续监控不确定性和证据覆盖率来规范诊断终止的观察者引导运行时控制 harness。

通过评估执行稳健性、推理动态以及诊断结果的严谨三层评估金字塔，EviDx 展现出了显著的性能和流程稳定性提升，同时也揭示了现有模型能力的边界。

摘要 (Summary)

Clinical diagnosis is inherently an active, iterative process where clinicians gather evidence, update competing hypotheses, and determine when sufficient information is available to make a definitive diagnosis. Traditional large language model (LLM) medical systems often treat diagnosis as a static, single-turn case-to-answer task with little capacity for dynamic evidence acquisition. While agentic LLMs offer dynamic tool use, they frequently lack precise runtime control and runtime scaffolding for patient evidence.

Clinical diagnosis is inherently an active, iterative process where clinicians gather evidence, update competing hypotheses, and determine when sufficient information is available to make a definitive diagnosis. Traditional large language model (LLM) medical systems often treat diagnosis as a static, single-turn case-to-answer task with little capacity for dynamic evidence acquisition. While agentic LLMs offer dynamic tool use, they frequently lack precise runtime control and runtime scaffolding for patient evidence.

EviDx 是一种新颖的证据感知主动诊断框架，它通过以下方式弥补了这一空白： 1. 通过 \(\mathcal{E}\)-Synthesis 从原始临床病例构建患者专属的诊断环境。 2. 建立一个组织角色专业化智能体、证据工具和演进中证据状态的临床诊断脚手架。 3. 引入一个观察者引导的运行时 harness，通过持续监测不确定性和证据覆盖率来规范诊断的终止。

EviDx is a novel evidence-aware active diagnosis framework that bridges this gap by combining: 1. Patient-specific diagnostic environments built from raw clinical cases via \(\mathcal{E}\)-Synthesis. 2. A clinical diagnostic scaffold organizing role-specialized agents, evidence tools, and evolving evidence states. 3. An observer-guided runtime harness that regulates diagnostic termination by continuously monitoring uncertainty and evidence coverage.

通过评估执行稳健性、推理动态和诊断结果的严谨三层评估金字塔，EviDx 证明了其在性能和过程稳定性方面的显著提升，同时也凸显了现有模型的客观能力边界。

Evaluated through a rigorous 3-level evaluation pyramid assessing execution robustness, reasoning dynamics, and diagnostic outcomes, EviDx demonstrates significant improvements in performance and process stability while highlighting existing model capability boundaries.

元数据 (Metadata)

摘要 (Abstract)

临床诊断是一个主动寻求证据的过程，在此过程中，临床医生获取证据、更新竞争假设，并决定何时现有的证据足以做出诊断。然而，许多围绕大语言模型（LLM）构建的医疗诊断系统仍然将诊断表述为静态的病例到答案预测，对证据获取的支持有限。尽管具备自主性的 LLM 通过工具使用和中间诊断轨迹提供了一种动态的替代方案，但现有系统往往对患者证据应如何在运行时进行公开、搭建脚手架和控制缺乏明确规范。我们介绍了 EviDx，这是一个证据感知的主动诊断框架，它将患者专属的诊断环境与临床诊断脚手架以及观察者引导的运行时 harness 相结合。在 EviDx 中，\(\mathcal{E}\)-Synthesis 从原始临床病例构建交互式环境；脚手架组织了角色专化的智能体、证据工具和演进的证据状态；而 harness 则通过追踪不确定性和证据覆盖率来规范诊断的终止。三层评估金字塔评估了执行稳健性、推理动态和诊断结果。实验表明，EviDx 提高了诊断性能和过程稳定性，同时揭示了依赖于模型的 capability 边界。

Clinical diagnosis is an active evidence-seeking process in which clinicians acquire evidence, update competing hypotheses, and decide when the available evidence is sufficient for diagnosis. Yet many medical diagnosis systems built around large language models (LLMs) still formulate diagnosis as static case-to-answer prediction, with limited support for evidence acquisition. Agentic LLMs offer a dynamic alternative through tool use and intermediate diagnostic trajectories, but existing systems often under-specify how patient evidence should be exposed, scaffolded, and controlled at runtime. We introduce EviDx, an evidence-aware active diagnosis framework that pairs patient-specific diagnostic environments with a clinical diagnostic scaffold and an observer-guided runtime harness. In EviDx, \(\mathcal{E}\)-Synthesis constructs interactive environments from raw clinical cases; the scaffold organizes role-specialized agents, evidence tools, and evolving evidence states; and the harness regulates diagnostic termination by tracking uncertainty and evidence coverage. A 3-level evaluation pyramid assesses execution robustness, reasoning dynamics, and diagnostic outcomes. Experiments show that EviDx improves diagnostic performance and process stability while revealing model-dependent capability boundaries.

访问与资源 (Access & Resources)