文章背景与核心概要

阿尔茨海默病（AD）的早期诊断对于及时干预、延缓病情发展及改善患者预后至关重要。传统的诊断方法往往具有侵入性或成本高昂，而基于语音的筛查由于非侵入性、成本低以及无需专业设备的特点，正受到越来越广泛的关注。然而，如何在利用大语言模型（LLM）强大语言表征能力的同时，保护高度敏感的患者语音及文本数据隐私，是当前临床落地的一大挑战。

本文介绍了一种名为 LSEAD 的创新型、非侵入式且具备隐私保护能力的语音分析框架。LSEAD 利用预训练的开源大语言模型在本地自动转录语音录音并提取丰富的语言文本嵌入，随后通过主成分分析（PCA）进行降维并用于分类。由于整个过程无需外部数据交换，该框架实现了极高的患者数据隐私保护。在 ADReSS20 和 ADReSSo2021 基准数据集上的评估表明，LSEAD 展现出强大的泛化能力，与现有方法相比，AD 分类准确率提升高达 5%，为真实世界临床环境中的大规模应用提供了一种高效且安全的解决方案。

LSEAD: A Privacy-Preserving LLM-Based Speech Analysis Framework for Early Alzheimer's Disease Screening

Summary

The paper introduces LSEAD, an innovative, non-invasive, and privacy-preserving framework designed for early Alzheimer's disease (AD) screening through speech analysis. Utilizing pretrained open-source large language models (LLMs), LSEAD automatically transcribes speech recordings and extracts rich linguistic text embeddings locally. By employing Principal Component Analysis (PCA) for dimensionality reduction prior to classification, the framework eliminates the need for external data exchange, ensuring high data privacy. Evaluated on the benchmark datasets ADReSS20 and ADReSSo2021, LSEAD demonstrates robust generalization and improves AD classification accuracy by up to 5% compared to existing methods, making it an efficient and scalable solution for real-world clinical environments.

Metadata & Document Information

元数据与文档信息

Authors

作者信息

• Xin Wang
• Yingchao Huang
• Yuhan Su
• Shanshan Yao
• Wei Peng

Abstract

摘要

早期诊断阿尔茨海默病（AD）对于实现及时干预至关重要，这可以减缓病情进展并改善患者预后。目前迫切需要非侵入性且具有成本效益的 AD 检测方法，特别是在面对具有不同患者群体和记录条件的真实世界临床环境时。基于语音的筛查通过使用无需专用设备收集的自然语音，满足了这些需求。大语言模型（LLM）的最新进展通过提供丰富的语言表征和强大的泛化能力，改善了语音分析。在这项研究中，我们提出了 LSEAD，这是一个使用预训练开源 LLM 的基于语音的 AD 检测框架。语音记录被自动转录，并使用本地部署的 LLM 提取文本嵌入。在分类前应用主成分分析（PCA）来降低维度。由于该框架仅依赖语音转录本和本地部署的模型，它支持无需外部数据交换的隐私保护型 AD 风险评估。我们在 ADReSS20 和 ADReSSo2021 基准数据集上评估了 LSEAD。实验结果表明，基于 LLM 的嵌入在不同数据集之间具有良好的泛化能力，与现有方法相比，AD 分类准确率提升了高达 5%，特别是在早期检测方面。这些结果表明，LSEAD 为早期 AD 筛查提供了一种实用、安全且可扩展的方法。

Early diagnosis of Alzheimer's disease (AD) is critical for enabling timely interventions that may slow disease progression and improve patient outcomes. There is a growing need for AD detection methods that are non-invasive and cost-effective, especially in real-world clinical settings with diverse patient populations and recording conditions. Speech-based screening addresses these needs by using natural speech collected without specialized equipment. Recent advances in large language models (LLMs) have improved speech analysis by providing rich linguistic representations and strong generalization. In this study, we propose LSEAD, a speech-based AD detection framework using pretrained open-source LLMs. Speech recordings are automatically transcribed, and text embeddings are extracted using locally deployed LLMs. Principal component analysis (PCA) is applied to reduce dimensionality before classification. Because the framework relies only on speech transcripts and locally deployed models, it supports privacy-preserving AD risk assessment without external data exchange. We evaluate LSEAD on the ADReSS20 and ADReSSo2021 benchmark datasets. Experimental results show that LLM-based embeddings generalize well across datasets and improve AD classification accuracy by up to 5 percent over existing methods, especially for early-stage detection. These results demonstrate that LSEAD provides a practical, secure, and scalable approach for early AD screening.

External Links & Resources

外部链接与资源

• Full-Text Access: View PDF | TeX Source
• External Indices:
• Google Scholar
• Semantic Scholar
• NASA ADS