跳转至

文章背景与核心概要

乳腺癌多学科团队（MDT）会议在面对日益复杂的病例和繁重的时间压力时，常常因繁重的文档记录需求而导致临床效率和决策质量下降。虽然基于AI的工作流可以缓解这一问题，但传统的云端处理由于需要处理可识别的患者数据，会带来严重的隐私风险。

为了克服这一挑战，研究人员开发了一个完全基于终端设备的边缘AI系统，该系统使用了开源的自动语音识别（ASR）和大语言模型（LLM）。该系统在单个 NVIDIA Jetson AGX Orin 上本地运行，能够安全地在机构基础设施内转录实时的MDT讨论、结构化临床信息，并利用基于英国国家健康与护理研究所（NICE）指南检索增强生成（RAG）的技术，生成循证治疗建议。

研究结果表明，经过优化的 Whisper模型在语音识别错误率上大幅降低，同时结合 MedGemma-RAG 技术能够比传统的云端专有比较对象识别出多出2.3倍的符合MDT共识的干预措施。这一成果展示了保护隐私、完全端侧运行的AI在MDT文档编写和指南驱动的决策支持方面的可行性，为未来的前瞻性临床评估打下了基础。

Development and Feasibility Evaluation of an Edge AI as Medical Device System for Breast Cancer Multidisciplinary Team Meetings

arXiv: 2608.22108 [cs.AI]
Submitted on: 22 August 2026
Primary Subject: Artificial Intelligence (cs.AI)
Secondary Subjects: Machine Learning (cs.LG)
Authors: Aarzoo Dhiman, Farzana Haque, Kartikae Grover, Lydia Brian Smith, William Stephen Jones

📌 Executive Summary

乳腺癌多学科团队（MDT）会议面临着日益增长的病例复杂性和时间限制，繁重的文档记录要求往往使这一情况雪上加霜。尽管AI驱动的工作流可以缓解这一问题，但传统的云端处理由于涉及可识别的患者数据，带来了严重的隐私风险。

为了解决这一问题，研究人员使用开源的自动语音识别（ASR）和大语言模型（LLMs），开发了一个完全在设备端运行的边缘AI系统。该系统在单个 NVIDIA Jetson AGX Orin 上本地运行，能够实现以下功能： * 在机构基础设施内部安全地转录实时的MDT讨论。 * 结构化临床信息。 * 利用基于英国国家健康与护理研究所（NICE）指南基础的检索增强生成（RAG）技术，生成循证治疗建议。

核心发现与性能

  • 转录优化： 对 Whisper large-v3 进行微调，使真实世界模拟录音的词错误率（WER）降低了 20.7%，结构化评估的WER降低了 24.4%，其性能几乎与商业临床ASR基准持平。
  • 临床准确性： MedGemma-RAG 识别出的符合MDT共识的干预措施是云端专有比较对象的 2.3 倍（\(p = 0.020\)），并在整体准确性上保持了一致性。
  • 利益相关者见解： 临床医生认为自动化文档、治疗决策支持和病例分诊是高价值的功能，同时指出无缝的工作流整合、严格的治理和临床医生的信任仍然是广泛应用的关键挑战。

Breast Cancer Multidisciplinary Team (MDT) meetings face growing case complexity and time constraints, often exacerbated by burdensome documentation requirements. While AI-driven workflows could alleviate this, traditional cloud-based processing creates severe privacy risks due to the handling of identifiable patient data.

To overcome this, the researchers developed a fully on-device Edge AI system using open-source Automatic Speech Recognition (ASR) and Large Language Models (LLMs). Operating locally on a single NVIDIA Jetson AGX Orin, the system: * Transcribes live MDT discussions securely within institutional infrastructure. * Structures clinical information. * Generates evidence-based treatment recommendations using Retrieval-Augmented Generation (RAG) grounded in National Institute for Health and Care Excellence (NICE) guidelines.

Key Findings & Performance

  • Transcription Optimization: Fine-tuning Whisper large-v3 reduced the word error rate (WER) by 20.7% on real-world simulated recordings and 24.4% on structured evaluations, achieving performance nearly matching commercial clinical ASR benchmarks.
  • Clinical Accuracy: MedGemma-RAG identified 2.3 times more MDT-concordant interventions than a proprietary cloud-based comparator (\(p = 0.020\)), maintaining parity in overall accuracy.
  • Stakeholder Insights: Clinicians highlighted automated documentation, treatment decision support, and case triage as high-value capabilities, while noting that seamless workflow integration, strict governance, and clinician trust remain pivotal challenges for widespread adoption.

📑 Abstract

乳腺癌多学科团队（MDT）会议在巨大的时间压力下处理着日益复杂的病例，而文档记录的要求可能会降低临床效率和决策质量。现有的基于AI的MDT工作流依赖于云端处理，由于患者讨论包含可识别的信息，这限制了它们的使用。我们开发了一个完全在设备端运行的AI流水线，利用开源的自动语音识别（ASR）和大语言模型（LLMs），能够转录乳腺癌MDT讨论、结构化临床信息，并利用基于英国国家健康与护理研究所（NICE）指南的检索增强生成（RAG）技术生成治疗建议。该流水线在单个 NVIDIA Jetson AGX Orin 上运行，确保患者音频、转录文本和输出结果保留在机构的基础设施内部。

评估包括两场录制的模拟MDT讨论、十场经过临床验证的合成讨论以及 1,270 条经过声学增强的录音。对 Whisper large-v3 的优化使得录制讨论的词错误率降低了 20.7% 和 24.4%，在增强音频上，其性能与商业临床ASR基准相比，WER差距在 0.58% 以内，词信息丢失率在 1.58% 以内。MedGemma-RAG 识别出的符合MDT共识的干预措施是云端专有比较对象的 2.3 倍（\(p = 0.020\)），整体准确性没有显著差异。利益相关者认为自动化文档、治疗建议支持和病例分诊是最具信服力的近期应用，同时也强调了工作流整合、治理和临床医生信任是关键的实施挑战。这些发现证明了保护隐私、完全端侧运行的AI在MDT文档和指南知情决策支持方面的可行性，为前瞻性临床评估奠定了基础。

Breast Cancer Multidisciplinary Team (MDT) meetings manage increasingly complex cases under considerable time pressure, and documentation requirements can reduce clinical efficiency and decision quality. Existing AI based MDT workflows rely on cloud-based processing, limiting their use because patient discussions contain identifiable information. We developed a fully on-device AI pipeline using open-source Automatic Speech Recognition (ASR) and Large Language Models (LLMs) that transcribes breast cancer MDT discussions, structures clinical information, and generates treatment recommendations using retrieval-augmented generation (RAG) grounded in National Institute for Health and Care Excellence (NICE) guidance. The pipeline runs on a single NVIDIA Jetson AGX Orin, ensuring that patient audio, transcripts, and outputs remain within institutional infrastructure.

Evaluation included two recorded simulated MDT discussions, ten clinically validated synthetic discussions, and 1,270 acoustically augmented recordings. Optimisation of Whisper large-v3 reduced word error rate by 20.7% and 24.4% on the recorded discussions and achieved performance within 0.58% WER and 1.58% word information lost of a commercial clinical ASR benchmark on augmented audio. MedGemma-RAG identified 2.3 times more MDT-concordant interventions than a proprietary cloud comparator (\(p = 0.020\)), with no significant difference in overall accuracy. Stakeholders identified automated documentation, treatment recommendation support, and case triage as the most credible near-term applications while highlighting workflow integration, governance, and clinician trust as key implementation challenges. These findings demonstrate the feasibility of privacy-preserving, fully on-device AI for MDT documentation and guideline-informed decision support, providing a foundation for prospective clinical evaluation.