MedPAO-Assist: Edge-Deployable, Modality-Aware Agentic AI for Radiology Report Structuring

Computer Science > Artificial Intelligence arXiv:2610.26913 (cs) [Submitted on 17 Oct 2026] Title: MedPAO-Assist: Edge-Deployable, Modality-Aware Agentic AI for Radiology Report Structuring Authors: Shrish Shrinath Vaidya, Minmini Selvam, Ganapathy Krishnamurthi.

计算机科学 > 人工智能 arXiv:2610.26913 (cs) [提交于 2026 年 10 月 17 日] 标题：MedPAO-Assist：用于放射科报告结构化的边缘部署、模态感知代理 AI。作者：Shrish Shrinath Vaidya, Minmini Selvam, Ganapathy Krishnamurthi。

Abstract: The adoption of Large Language Models for radiology report structuring has been largely dominated by either proprietary cloud-based models or large open models requiring high-end GPUs for deployment, which is a critical bottleneck for clinical adoption in radiology workflows that demand data privacy and offline operation.

摘要：大型语言模型在放射科报告结构化方面的应用，目前主要被专有的云端模型或需要高端 GPU 部署的大型开源模型所主导，这对于要求数据隐私和离线操作的放射科临床工作流程来说，是一个关键的瓶颈。

To address this, we present MedPAO-Assist, an edge-deployable agentic framework that generates structured radiology reports directly from radiologist dictation and voice commands, requiring only $\approx3.3$ GB peak RAM and achieving an average end-to-end latency of $\approx31$ seconds on a standard laptop CPU through an asynchronous, multi-threaded execution pipeline.

为了解决这一问题，我们提出了 MedPAO-Assist，这是一个可边缘部署的代理框架。它能直接根据放射科医生的口述和语音指令生成结构化放射科报告，仅需约 3.3 GB 的峰值内存，并通过异步多线程执行流水线，在标准笔记本电脑 CPU 上实现了约 31 秒的平均端到端延迟。

The system currently supports four major imaging modalities — Chest X-ray, Chest CT, Abdomen CT, and Brain MRI. To our knowledge, it is among the earliest agentic systems to run its entire pipeline with a Small Language Model (SLM) in the 0.6B-parameter range.

该系统目前支持四种主要的影像模态——胸部 X 光、胸部 CT、腹部 CT 和脑部 MRI。据我们所知，它是最早使用 0.6B 参数范围的小型语言模型（SLM）运行整个流水线的代理系统之一。

This multi-agent architecture, built around an Extract-Ground-Generate paradigm, performs concept extraction, SNOMED-CT grounding, and structured report generation, followed by a speech-driven human-in-the-loop correction phase, together forming a fully “no-touch” reporting workflow that requires no keyboard input, external API calls, or internet connectivity.

这种围绕“提取-基础-生成”（Extract-Ground-Generate）范式构建的多代理架构，执行概念提取、SNOMED-CT 基础映射和结构化报告生成，随后进行语音驱动的人机协同校正阶段，共同形成了一个完全“零接触”的报告工作流程，无需键盘输入、外部 API 调用或互联网连接。

We evaluate each pipeline stage individually — concept extraction, ontology grounding, and concept categorization and report hardware profiling results demonstrating the system’s feasibility for real-time, CPU-only clinical deployment.

我们分别评估了流水线的每个阶段——概念提取、本体基础映射和概念分类，并提供了报告硬件分析结果，证明了该系统在仅使用 CPU 的情况下进行实时临床部署的可行性。