早期阿尔茨海默症的言语特征变异性及其神经机制研究

As the most common type of senile dementia, Alzheimer's disease (AD) shows progressive and irreversible damage to cognitive function. Early detection of AD is benefit to provide appropriate cognitive interventions to patients and delay the process of cognitive impairment. Language deficit, one of the important clinical symptoms of preclinical and early AD, is significantly related to various development stages of AD course of disease. Therefore, the objective measures of the speech deficit of early AD in different language domains will provide significant information for AD early detection and diagnosis. However, high-sensitivity, high-specificity speech and language deficit characteristics that can reflect neural information transmission deficits in AD are not available. The proposed project will intend language deficiency of early AD with the following aspects from the point of view of behavioral and neural mechanism: 1) Feature extraction methods of early AD patients’ abnormal speech characteristics in Chinese mandarin language; 2) Exploring the lesion changes in the dynamic reorganization of large-scale language functional connectivity networks on speech production in early AD patients; 3) Study the relevancy between damage features of brain language functional network and abnormal speech characteristics of AD patients, and then obtain the speech feature set sensitive to early AD; 4) Establish classifying mode with multi-task deep learning and achieve the recognition to early AD patients. The research achievements not only will provide crucial academic value to the research of neural mechanism of the lesion of language functional connectivity network due to Alzheimer's disease, but will contribute the important technology and theoretical basis to early screening of AD patients by using non-invasive, telemedicine and effective approach. It has important academic value and good prospect for clinical application.

阿尔兹海默病（AD）是老年痴呆最常见的类型，认知功能呈进行性且不可逆性损害。AD早期发现有助于对患者进行早期干预，延缓认知损害进程。语言障碍是早期AD的重要临床表征之一，且与AD病程发展各阶段密切相关。因此，分析AD语音的变异信息将为AD早期筛查、诊断提供重要信息。然而，当前尚未找到高敏感性、高特异性地反映AD语言神经信息传导损伤的语音特征集。本课题拟从行为与脑机制角度，探究：1）中国普通话语种早期AD患者异常语音特征的提取方法；2）AD患者脑语言功能网络损伤影响言语产生的机制；3）探索AD的脑语言功能网络损伤特征与异常语音特征的关联性，获得对AD敏感的语音特征集；4）建立多任务深度学习的分类模型，实现对早期AD患者的识别。项目的研究成果不仅对AD患者脑语言功能网络损伤的神经机制研究有重要的学术价值，且为远程、大范围AD患者早期筛查提供一种重要技术手段和理论依据，具有良好的临床应用前景。

阿尔兹海默病（AD）是老年痴呆最常见的类型，认知功能呈进行性且不可逆性损害。AD早期发现有助于对患者进行早期干预，延缓认知损害进程。语言障碍是早期AD的重要临床表征之一，且与AD病程发展各阶段密切相关。因此，分析AD语音的变异信息将为AD早期筛查、诊断提供重要信息。然而，当前尚未找到高敏感性、高特异性地反映AD语言神经信息传导损伤的语音特征集。因此，本项目拟从行为和脑功能的角度出发，探索AD患者对言语加工过程的影响以及敏感性指标，进一步建模，构建了基于语音分析的AD早期筛查预警系统，并且这一成果已经被华为公司作为运动健康的附加产品进行了进一步的项目合作，已签订项目计划书。本项目在国家自然基金的资助下，主要完成了以下主要内容：1）建立轻度AD（MCI患者）和认知正常老年人多模态、多任务数据库，包括多任务的言语产生的语音数据库、言语产出过程的动态脑电数据库和神经心理学量表数据库；2）构建言语产生认知加工过程中脑网络的动态功能连接模型，揭示MCI患者言语功能障碍的产生机制；3）研究了基于脑功能影像方法和深度学习融合对老年退行性疾病诊断的方法，实现对帕金森症，AD患者的诊断和治疗预测方法；4）完成了基于多任务言语功能评估方法、MCI自发语音的韵律特征提取方法和基于多任务深度学习的建模方法等，构建了MCI认知障碍的语音评估模型，可以有效的、自动对MCI患者进行筛查，筛查准确率不低于90%，从而为开发可用于普通话的MCI认知障碍的早期自动筛查系统，使之能为认知障碍的早期筛查方法，提供快速、非侵入性的辅助诊断依据打下良好的研究基础；5）进一步，研究了非侵入式脑功能刺激手段对于老年退行性疾病的治疗效果。本研究的成果可用来指导未来的包括认知障碍的早期筛查和康复治疗的干预性治疗手段的改进。