基于功能磁共振成像的脑网络研究

发布时间：2018-08-12 11:16

【摘要】：人脑拥有1000亿个神经元和100万亿个神经突触,是宇宙中已知的最为复杂的系统。多时空尺度的信息调和是脑功能支撑所有认知和行为的先决条件。因此,为了系统全面准确地探索大脑的工作机制,脑科学与认知神经科学研究须从连接和网络的层面进行。功能磁共振成像(functional magnetic resonance imaging,f MRI)作为一种非侵入性活体脑功能成像方法,具有较高的时空分辨率,为研究大尺度人脑功能网络提供了强有力的手段。本文以静息态f MRI为主要成像方法,进行了自发头动的神经网络机制和自发神经活动的时空动态特性等基础问题研究,以及基于功能连接磁共振成像(functional connectivity MRI,fc MRI)的脑状态有监督和无监督模式分析的方法学和相关应用问题研究。论文工作主要包括以下四个方面的内容:脑成像过程中自发头动的神经网络机制。脑指标的个体差异,特别是采用f MRI度量的功能连接,与数据采集过程中的头动差异具有相关性。这一结果的前提是,头动伪迹引起的连接差异被完全校正。本文第二章提出,脑连接差异也可能是预测头动差异的神经生物学特性。我们采用个体间和个体内的大头动数据和小头动数据的比较分析来支持这一论述。个体间分析找到了头动的生物学标记,即大头动被试默认网络内的远程功能连接减弱。在个体内的分析中则没有观察到相似的连接差异。相反,这一生物学标记在个体内具有很好的稳定性。这些发现表明,脑连接的差异不能简单地认为是头动伪迹效应,也有可能反映的是功能组织的个体差异。将疾病的生物学标记从本文报道的头动生物学标记区分开,可能对于确定治疗靶标和改进治疗方案具有重要意义。基于全脑静息fc MRI的重症抑郁症鉴别。本文第三章论证了基于全脑静息fc MRI的有监督模式分析方法鉴别重症抑郁症患者的可行性,结果证明基于全脑静息fc MRI的有监督分类方法可以以94.3%的分类精度将重症抑郁症患者从健康对照中鉴别出来。最具有区分能力的功能连接主要分布在默认网络、情感网络、视皮层区域和小脑内部或之间,表明这些与疾病相关的静息网络可能是引起重症抑郁症患者情感和认知紊乱的重要原因。此外,杏仁核、前扣带皮层、海马旁回和海马可能在抑郁症病理中扮演了重要角色。这些结果表明全脑静息fc MRI可以作为抑郁症诊断的潜在生物学标记。基于静息fc MRI的无监督脑网络模式分类框架。当前的精神疾病包括抑郁症的诊断都是基于病人自述症状及临床表现的,因此可能会受病人的当前行为和医生的偏见的影响。本文第四章提出基于静息fc MRI的无监督脑网络模式分类框架,在不利用诊断类别标签信息的情况下,采用fc MRI数据的无监督聚类分析鉴别抑郁症。我们采用静息功能连接在膝部扣带皮层内定义了两个亚区:膝下区和膝前区。基于膝下区的功能连接图谱,最大间隔聚类算法可以将抑郁症患者从健康被试中鉴别出来,组水平聚类正确率和个体水平分类正确率均达到92.5%。此外,最具区分能力的膝下扣带功能连接网络包括腹外侧和腹内侧前额叶、颞上回和边缘区域,这些连接可能在抑郁症的病理中扮演了重要角色。该研究表明,膝下扣带功能连接网络特性可以为重症抑郁症的诊断提供理想的客观的生物学标记,也表明基于最大间隔聚类的无监督机器学习方法在临床实践和辅助精神疾病研究中的潜能。人脑自发神经活动的时空动态特性。默认网络和注意网络在多个维度上的对立可能暗示了大脑组织的一个特征,即目标竞争的认知处理不仅在空间上是分离的,在时间上也可能是分离的。本文第五章通过单个BOLD(blood oxygenation level dependent,血氧水平依赖)-f MRI时间相位的共活动模式聚类得到了脑状态的瞬时合成与分解,显示了大脑活动在时域上由多个相互竞争的空间翻转模式所构成。这些活动模式与经典的功能网络关系是不一样的,并且在脑区内部可以在时域上动态地遵循截然不同的组织边界和功能边界。脑活动状态间的迁移是高度对称的,特别地,对立脑状态间的迁移概率是极低的。在执行语义分类任务时,这些空间模式的时域组织规律是相似的,与任务相关的状态在静息态下也可以观察得到,只是在任务态下有选择性地出现而已。总之,这些结果表明,在静息状态下,人脑活动在时域上表现为动态地在固定的功能模式剧目间切换,而这个剧目是由一些对立的活动模式对所主导的。通过提高对大脑固定的功能剧目属性的认识(包括确定它们潜在的电生理相关性),瞬时状态可能更好地揭示持续进行的认知处理过程,提供一个行为变化的指标,并作为疾病的潜在的生物学标记。
[Abstract]:The human brain, with 100 billion neurons and 100 billion synapses, is the most complex system known in the universe. Information reconciliation at multiple spatial and temporal scales is a prerequisite for brain function to support all cognition and behavior. Functional magnetic resonance imaging (f MRI), as a non-invasive in vivo brain functional imaging method, has a high spatial-temporal resolution and provides a powerful means for the study of large-scale human brain functional networks. The basic research on the neural network mechanism of head movement and the spatiotemporal dynamic characteristics of spontaneous nerve activity, as well as the methodologies and related applications of supervised and unsupervised mode analysis of brain state based on functional connectivity MRI (fc MRI). The work of this paper mainly includes the following four aspects Rong: Neural network mechanism of spontaneous head movement in brain imaging. Individual differences in brain indices, especially functional connectivity measured by F MRI, are correlated with head movement differences in data acquisition. The premise of this result is that the connection differences caused by head motion artifacts are corrected completely. In the second chapter of this paper, it is proposed that brain connection differences may also be possible. It is a neurobiological feature that predicts head movement differences. We support this argument by comparing the large head motion data and the small head motion data between individuals and individuals. Inter-individual analysis found the biological marker of head movement, i.e. the weakening of long-range functional connectivity in the default network of the large head motion subjects. In intra-individual analysis, this was not the case. Similar junctional differences were observed. On the contrary, this biomarker was very stable in individuals. These findings suggest that differences in brain junctions can not be simply considered as head-movement artifacts, but may also reflect individual differences in functional tissues. The third chapter demonstrates the feasibility of supervised pattern analysis based on whole brain resting FC MRI in differentiating patients with severe depression. The results demonstrate the supervised classification based on whole brain resting FC MRI. The most discriminatory functional connections were found in the default network, the emotional network, the visual cortex and the cerebellum, indicating that these resting networks associated with disease may be responsible for the emotional and cognitive development of patients with major depression. In addition, the amygdala, anterior cingulate cortex, parahippocampal gyrus and hippocampus may play important roles in the pathology of depression. These results suggest that resting FC MRI may be a potential biological marker for the diagnosis of depression. The diagnosis of depression is based on the patient's self-reported symptoms and clinical manifestations, so it may be influenced by the patient's current behavior and doctor's prejudice. In chapter 4, we propose an unsupervised brain network pattern classification framework based on resting FC MRI, which uses unsupervised clustering of FC MRI data without using the label information of diagnostic categories. We defined two sub-regions in the knee cingulate cortex: the subgenu and the anterior genu. Based on the functional connectivity map of the subgenu, the maximum interval clustering algorithm can distinguish the depressive patients from the healthy subjects. Both the group level clustering accuracy and the individual level classification accuracy are achieved. 92.5%. In addition, the most differentiated subgenicular cingulate functional connectivity networks include the ventrolateral and ventromedial prefrontal lobes, superior temporal gyrus, and marginal regions, which may play an important role in the pathology of depression. This study suggests that the subgenicular cingulate functional connectivity network characteristics can provide ideal and objective information for the diagnosis of major depression. Biological markers also indicate the potential of unsupervised machine learning based on maximum-interval clustering in clinical practice and in assistant psychiatric research. Spatiotemporal dynamics of spontaneous neural activity in the human brain. In the fifth chapter, the instantaneous synthesis and decomposition of brain states are obtained by clustering the co-activity patterns of a single BOLD (blood oxygenation level dependence) and the time phase of F MRI, showing that brain activities compete with each other in the time domain. These patterns are different from the classical functional networks and can dynamically follow distinct organizational and functional boundaries within the brain area in time domain. When performing semantic classification tasks, the temporal organization of these spatial patterns is similar. Task-related states can be observed in resting state, but selectively appear in task state. In short, these results indicate that human brain activities in resting state are dynamically fixed in the temporal domain. Transient states may better reveal ongoing cognitive processing, provide an indicator of behavioral change, and serve as a guide to behavioral change by increasing awareness of the brain's fixed functional repertoire attributes (including determining their potential electrophysiological correlations). Potential biomarkers of disease.
【学位授予单位】：国防科学技术大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R445.2;R338

【共引文献】