随着年龄的增长，认知功能会随着老龄化的发展而发生衰退，而认知功能对人类的正常生活有着极为重要的作用。认知老化的深入研究，包括认知老化的发展规律、对认知老化的预测及影响因素等研究，特别是认知老化发展的预测的相关研究，对人类认知功能的发展及以认知功能障碍为主要疾病特征的老年痴呆等神经退行性疾病的预防、早期发现、及时干预及追踪有深刻的意义。 大脑是人类一切行为的基础，多个脑区的相互作用构成的大脑结构网络是完成各项认知功能的结构基础，弥散张量成像技术的发展及基于图论的复杂网络分析方法可以帮助我们揭示大脑结构网络的拓扑属性。而以往研究更多的是探究大脑白质完整性以及大脑功能网络与认知老化的关系，较少探究大脑结构网络与纵向认知变化之间的关系，特别缺乏大脑结构网络对老年人纵向认知变化预测的研究。而本研究就基于图论分析，以老年人的大脑结构网络拓扑属性与纵向认知变化为研究对象，探究大脑结构网络与纵向认知变化的关系及预测作用。 本研究数据来源于北京老年脑健康计划（Beijing aging brain rejuvenation initiative，BABRI）数据库，选取纵向追踪认知测验及核磁数据完整的被试，共90人。被试所接受的认知功能测验主要包括一般认知能力（简易智能精神状态量表），注意功能（符号数字转换测验，连线测验A部分），记忆功能（Rey听觉词语记忆测验，Rey-Osterrich复杂图形测验延迟回忆，数字广度测验），执行功能（连线测验B部分，Stroop色词测验）和言语能力（词语流畅性测验，波士顿命名测验），并计算各项认知测验的月变化率。MRI扫描采集的影像包括磁共振扫描T1加权结构像和DTI扫描。基于图论分析的方法获取每个被试大脑结构网络拓扑属性（全局效率、局部效率、节点全局效率及节点局部效率），基于多核支持向量回归方法纳入大脑结构网络指标及人口学指标为特征探究大脑结构网络对认知变化的预测作用。 认知测验纵向变化与大脑结构网络的偏相关分析结果表明，数字符号测验、连线测验和命名测验的月变化率与基线时大脑结构网络的全局效率和局部效率之间有显著的相关关系；进一步的预测分析说明，大脑结构网络节点效率与人口学指标为特征的多核支持向量回归SVR模型能较好的预测认知测验的纵向变化，并且多核SVR的效果也要显著好于仅使用大脑结构网络指标或人口学指标的单核SVR；对预测中重要特征的分析结果表明，在注意功能测验的预测关键脑区是左侧顶上回和缘上回，执行功能预测关键脑区是左侧后扣带和楔叶，而言语功能的预测关键脑区是右侧颞中回、颞下回和中央前回。 本研究采用多核支持向量回归的方法，发现了基线时大脑结构网络指标与老年人注意功能、执行功能、言语功能的纵向认知变化的相关关系以及预测作用，未来的研究可以纳入更多的大脑属性及异常老化人群探究不同老化中早期大脑属性对纵向认知变化的预测作用。

Cognitive function declines with the development of aging, and cognitive function has a very important role in the normal life of human beings. An in-depth study of cognitive aging, including the development of cognitive aging, predictions of cognitive aging, and influencing factors, especially studies related to the prediction of cognitive aging, has great significance on the development of human cognitive function and the prevention, early detection, timely intervention, and tracking of neurodegenerative diseases such as dementia with cognitive impairment. The brain is the basis of all human behavior. The brain structure network formed by the interaction of multiple brain regions is the structural basis for completing various cognitive functions. The development of diffusion tensor imaging technology and complex network analysis methods based on graph theory can help us reveal the topology properties of the brain structure network. In the past, more studies focused on the relationship between the white matter integrity of the brain, the brain function network and cognitive aging, and less on the relationship between the brain structure network and the longitudinal cognitive changes, especially the study of the brain structure network predicts cognitive changes for the elderly. This research is based on graph theory analysis, and takes the topological properties of the brain structure of the elderly and longitudinal cognitive changes as the research object, and explores the relationship between the brain structure network and longitudinal cognitive changes and predictive function. The data of this study were collected from the Beijing aging brain rejuvenation initiative (BABRI) database. A total of 90 participants were selected for the longitudinal tracking cognition test and complete magnetic resonance imaging data. The cognitive function tests accepted by the participants included general cognitive ability (Mini Mental Status Examination，MMSE), attentional function (Symbol Digit Modalities Test, Trail Making Test A, Stroop Color-Word Test B), and memory function (Rey Auditory Verbal Learning Test, Rey-Osterrich Complex Figure Test, Digit Span Test), Executive Function (Trail Making Test Part B, Stroop Color-Word Test) and Language Ability (Boston Naming Test, Verbal Fluency Test), and calculated monthly changes rate in cognitive tests. Images acquired on MRI scans included magnetic resonance scan T1 weighted structural images and DTI scans. Based on the graph theory analysis method, the network topological properties (global efficiency, local efficiency, node global efficiency, and node local efficiency) of each tested brain structure were obtained. Based on multi-kernel support vector regression method, the characteristics of brain structure network index and demographic index and the prediction of cognitive changes by the brain structure network were explored. The partial correlation analysis between the longitudinal changes of the cognitive test and the brain structure network showed that there was a significant difference between the monthly rate of change of the Symbol Digit Modalities Test, Trail Making Test B, and the Boston Naming Test and the global efficiency and the local efficiency in the brain structure network at the baseline. Further predictive analysis showed that the multi-kernel support vector regression method (SVR) model characterized by node efficiency of the brain structure and demographic indicators can better predict the longitudinal changes of cognitive tests, and the effect of multi-kernel SVR is also significantly better than only Single-core SVR using brain structure network indicators or demographic indicators; Analysis of important features in predictions showed that the key brain regions predicted the attention function are the left Superior occipital gyrus, the left Superior parietal gyrus and supraorbital gyrus. The key brain regions of predicting executive function were the left posterior cingulate and cuneate, and the key brain regions predicted language function were the right middle temporal gyrus, the right inferior temporal gyrus, and the central anterior gyrus. This study used multi-kernel support vector regression method (SVR) model to find out the correlation between the brain structure network index and the longitudinal cognitive changes of attention function, executive function, and language function of the elderly and the predictive role. The future research can incorporate more brain properties and abnormal ageing populations to explore the predictive effects of different brain properties in early aging on longitudinal cognitive changes.