基于静息功能磁共振成像的人脑静息态网络研究是当前认知神经科学领域和神经精神疾病领域的重要研究方向之一。本文将主要在系统层面探讨静息态网络间内在的功能整合规律和连接模式，并应用到阿尔茨海默病研究中，从静息态网络内和网络间全局整合角度理解其认知障碍神经机制，挖掘一些对疾病敏感和特异的影像学生物指标。本研究共收集三组静息态功能磁共振成像数据，包括健康年轻人组、阿尔茨海默病人组和正常老年控制组数据，采用“从分离到整合”的研究方式，首先通过独立成分分析方法从静息功能像数据中分离静息态网络，然后运用贝叶斯网络学习方法，将静息态网络与贝叶斯网络节点关联，构建网络间连接模型。主要研究内容是：一、在计算方法学上，完善贝叶斯网络学习方法在构建静息态有效连接模型中的若干问题，包括贝叶斯网络的统计检验和贝叶斯网络组间比较等问题，探索研究静息态下神经系统间如何进行信息处理的动力学机制的新方法；二、在认知上，构建正常人脑静息态网络间的功能整合和信息传递模式，探讨分离的静息态网络如何组织起来以及静息态网络间如何进行信息处理；三、在应用方面，研究阿尔茨海默病对静息态网络及其关系的影响规律，建立一些有潜在应用价值的生物指标。本文研究发现，在正常人脑中静息态网络按照其功能层次属性（低层次感觉网络和高层次认知网络）以等级方式进行组织；从感觉网络到认知网络的自下而上的连接模式可能主导静息态网络间的信息传递；感觉网络和认知网络之间存在一些负连接关系，可能体现了二者之间内在的对大脑资源的竞争或者大脑资源的动态分配；默认网络作为一个重要的汇聚网络接收来自其它网络的信息，并对始于感觉网络的自下而上的信息进行最终整合。在阿尔茨海默病状态下的大脑中，网络内功能连接的衰退伴随着更为广泛的网络间大尺度连接的变异，注意、自省和默认网络等与高层认知加工有关的网络其功能连接和网络间的有效连接受到疾病的显著影响，视觉、听觉、感觉-运动等感觉网络的功能连接和之间的有效连接受到的影响则较少；静息态网络间连接衰退的同时可能会出现连接补偿和连接随机化；阿尔茨海默病改变了静息态网络间的组织模式，抑制了默认网络对其他网络的功能整合能力；默认网络内后扣带回和内侧前额叶，以及背侧注意网络左侧额叶眼区和顶内沟等区域的静息活动程度对疾病有高度的敏感度和特异度。总之，本文以“从分离到整合”的角度探讨了静息态下神经系统间的功能整合和连接模式问题，丰富了目前关于大脑自发性神经活动动力学机制的相关研究；在阿尔茨海默病研究中，将传统的单一静息态网络分析扩展到更多网络及其关系研究，加深了关于阿尔茨海默病“失连接综合症”理论的理解；本文联合的数据驱动的独立成分分析和贝叶斯网络学习方法为探讨静息态下功能网络的分离和整合机制提供了新的研究途径。

Using resting-state functional magnetic resonance imaging (fMRI) technique to examine the resting-state networks (RSNs) in human brain has been one of the focuses of present cognitive neuroscience and neuropsychiatric disorder studies. This work intends to offer our view on the global organizational properties and the inter-network connectivity patterns of these RSNs, and apply these studies to the exploration of Alzheimer’s disease (AD), aiming to understand its cognitive disorder mechanism in the view of overall connectivity and establish some sensitive and specific biomarkers.Three resting-state fMRI datasets including healthy young group, AD group and normal elderly control group (NC) were involved in this study. By first separating RSNs using Group independent component analysis (ICA) and then Bayesian network (BN) learning for the modelling of directional inter-network connectivity patterns, we analyzed each dataset from separation to integration. The main works are: 1) in methodology, address some issues on the application of BN in resting-state effective connectivity constructions, including statistical testing of BN model and between-BN comparisons, and establish new data analysis procedure for the study of resting-state connectivity patterns; 2) in cognitive, construct the global organizational properties of these RSNs and delineate the cross-network information processing patterns of resting human brain, and explore how these RSNs organize together and how information exchange occur among these RSNs during resting-state; 3) apply above studies to the exploration of AD, evaluate the AD-induced overall inter-network connectivity alterations, and establish some potential biomarkers.BN-base RSNs interconnectivity shows that networks at different cognitive function levels were hierarchically interconnected and organized, and constant bottom-up (from sensation RSN to cognitive RSN) cross-network information exchanges might dominantly and intrinsically engaged in the spontaneous neural activity. Some connections with negative weights were found particularly for the direct connectivity between sensory networks and cognitive networks, indicating a brain resource competition and dynamic reallocation between them. Among the information exchanges, the default-mode network (DMN) may be pivotal in integrating the resting-state information. In AD, declined within network functional connectivity was accompanied by more extensive large-scale inter-network connectivity alterations. Connections related with higher cognitive network such as attention, self-referential and the DMN, were more vulnerable to be impaired by the disease comparing with sensory networks such as visual, auditory, sensory-motor network. The results also suggested loss of connections in AD accompanied network connectivity compensation and randomization. The important DMN also decreased its information integration functionality among these networks in AD. We also found activity in the left intraparietal sulcus and left frontal eye field from dorsal attention network as well as the posterior cingulate cortex and medial prefrontal cortex from the DMN could serve as sensitive and specific biomarkers distinguishing AD from NC.In short, this work investigated the global organizational properties of RSNs and the large-scale cross-network connectivity patterns in the perspective of “from separation to integration”, which would enrich present studies on spontaneous neural activity dynamics. In AD, we expanded traditional individual network alteration examinations to more networks and their relations studies, which would deepen our understanding of the “disconnection syndrome” in AD. The combined data-driven method “ICA+BN” provide new avenue for the exploration of RSNs separation and integration.