基于多模态磁共振成像的儿童抽动秽语综合征脑结构和功能变化及计算机辅助诊断研究

CASIA OpenIR > 毕业生 > 博士学位论文

	基于多模态磁共振成像的儿童抽动秽语综合征脑结构和功能变化及计算机辅助诊断研究
	文宏伟1,2
	2017-05-31
学位类型	工学博士
中文摘要	抽动秽语综合征(Tourette syndrome, TS)是一种儿童常见的神经精神性疾病，TS的病因至今尚不明确，临床缺乏客观指标评价TS及估计预后。TS复杂的临床表现使得其诊断仍具有相当的主观性。本文基于多模态磁共振成像，从脑组织的微观结构、脑网络的宏观拓扑组织层面，对TS患儿大脑的结构和功能异常改变进行全面研究，并结合模式识别与机器学习技术实现TS患儿和正常儿童的自动化准确分类，为辅助TS的临床诊断提供了客观的神经影像学生物标志，最后基于深度学习技术实现脑结构网络到功能网络的预测，揭示了TS患儿脑网络结构和功能耦合性的改变。研究内容具有重要的理论意义和临床价值。本文主要研究内容及创新点如下：（1）我们提出了结合体素尺度上的基于纤维束示踪的空间统计分析和感兴趣区尺度上的基于大变形微分同胚尺度映射的脑图谱分析方法，对TS患儿的脑微观结构异常改变进行综合研究。研究揭示了TS患儿的脑组织微观异常不局限于运动通路，躯体感觉通路、边缘系统等同样受累。（2）在TS患儿脑结构和功能网络宏观拓扑组织的研究中，我们利用图论方法，在多稀疏度阈值下对TS患儿脑结构和功能网络的全局拓扑属性和节点拓扑属性的异常改变进行了综合研究。结果表明TS患儿的结构网络和功能网络的全局、局部效率明显降低，而最短路径长度和小世界属性明显增大。此外，TS患儿的结构网络和功能网络中节点拓扑属性发生显著改变的区域不仅包括感觉运动区域，还包括视觉功能、默认模式网络和语言功能相关的区域。相比于结构变化，功能变化程度更为显著，且功能网络和结构网络呈现不同的核心节点分布和拓扑异常区域。研究结果表明脑功能连接受结构连接的影响。（3）在基于脑组织微观结构特征的TS分类研究中，我们提出了一种基于多核学习方法融合多模态多类型特征的自动分类框架，并取得了该领域目前最高的分类正确率（94.24%）。我们还发现具有高区分能力的特征主要位于皮层-基底核环路，额叶皮层-纹状体-丘脑环路，这些区域和TS的病理高度相关。本研究对辅助TS临床诊断及病理学研究提供了潜在的神经影像学的生物标志。（4）在基于脑功能和结构网络宏观拓扑特征的TS分类研究中，我们提出了一种改进的相似网络融合方法，将不同稀疏阈值的网络视为不同类型特征表达，并分别构建了多阈值融合的功能和结构网络。结果表明相似网络融合方法极大地提升了网络拓扑特征的组间可区分性和分类准确率，在功能和结构网络的实验中分别取得了88.79%和89.13%的高准确率。（5）在脑网络结构到功能的预测研究中，我们提出了一种基于深度典型相关自编码(deep canonically correlated autoencoders, DCCAE)的计算模型。结果显示基于深层结构的DCCAE去建模复杂的多模态数据关联时相比浅层结构更为有效，明显提升了脑网络结构到功能预测的准确性。与此同时，我们发现TS患儿尽管保留了与健康儿童高相似性的功能模块划分，但结构和功能耦合性在全脑水平及运动感觉、默认网络、皮层下模块内均显著性增大，这表明TS患儿脑结构对功能的约束较健康儿童更强，使得功能缺乏动态性和灵活性。
英文摘要	Tourette syndrome (TS) is a childhood-onset neurological disorder. To date, the neural basis of TS remains unknown, and accurate TS diagnosis remains challenging due to its varied clinical expressions and dependency on qualitative description of symptoms. In this dissertation, we used multi-modal MRI to investigate the brain structural and functional alterations in TS children. We also applied pattern recognition and machine learning to identify objective neuroimaging biomarkers which may help improve early TS diagnosis. The main contents and contributions of this dissertation are as the follows: (1) We combined Tract-Based Spatial Statistics (TBSS) and atlas-based regions of interest (ROI) analysis, to investigate the microstructural diffusion changes in both deep and superficial white matter (WM) in TS children. We found widespread changes with significant decreased fractional anisotropy (FA) and increased radial diffusivity (RD). Furthermore, we found that lower FA values and higher RD values in WM regions are significantly correlated with more severe tics. Our results indicate that TS is not restricted to motor pathways alone, but affects somatosensory pathways, limbic structures. (2) We used diffusion MRI probabilistic tractography and resting-state functional connectivity (FC) to construct the structural and functional networks, respectively. We applied graph theoretical analysis to investigate the topological organization of structural and functional networks at both global and regional level. Compared to controls, TS children exhibited decreased global and local efficiency, increased shortest path length and small-worldness, indicating a disrupted balance between local specialization and global integration in structural and functional networks. In addition, TS children exhibited significant altered nodal topological properties in structural and functional networks, mainly distributed in the sensorimotor, default-mode, language and visual association areas. However, the topological alterations of functional networks were much more significant than structural networks. Moreover, the hub distribution and disrupted regions in structural and functional networks were also different, indicating that FC is likely constrained by structural connectivity (SC). (3) We proposed a classification framework based on multiple kernel learning (MKL) to fuse the features extracted from multimodal MRI. We combined TBSS, VBM and atlas-based ROI analysis to extract five types of features, and then each kind of feature is assigned a different kernel; finally a MKL classifier is trained and each kernel is assigned a weight to fuse all features. We achieved the highest classification accuracy of 94.24% and the most discriminative brain regions were mostly located in the cortico-basal ganglia, frontal-striatal-thalamic circuits, highly related to TS pathology. Our method might provide helpful neuroimaging biomarkers for assisting the clinical TS diagnosis. (4) We investigated the brain structural and functional networks using machine learning methods, to assist clinical TS diagnosis. We considered the structural and functional networks with different sparsity thresholds as diverse feature types, and adopted the similarity network fusion algorithm to merge the multi-threshold networks for individual subject. Then we extracted the topological properties of networks as features for TS classification. The results showed that the fused network incorporates the complementary information from multi-threshold networks, so it can better represent the underlying high-level structure of the original network. We achieved high classification accuracy (88.79% and 89.13%) using the fused structural and functional networks, respectively. (5) We proposed a computational model to accurately predict human resting-state FC from SC based on deep canonically correlated autoencoders (DCCAE). We got better predicting performance using deep model than shallow model. Although five common functional modules were identified in both groups, the patients showed the significantly increased SC-FC coupling at levels of whole brain and sensorimotor, default-mode, subcortical modules. The increased SC-FC coupling may suggest that TS pathology leads to functional interactions that are more directly related to the underlying SC of the brain and may be indicative of more stringent and less dynamic brain function in TS children.
关键词	抽动秽语综合征脑结构与功能多核学习相似网络融合深度典型相关自编码
学科领域	模式识别与智能系统
文献类型	学位论文
条目标识符	http://ir.ia.ac.cn/handle/173211/14652
专题	毕业生_博士学位论文
作者单位	1.中国科学院自动化研究所 2.中国科学院大学
第一作者单位	中国科学院自动化研究所
推荐引用方式 GB/T 7714	文宏伟. 基于多模态磁共振成像的儿童抽动秽语综合征脑结构和功能变化及计算机辅助诊断研究[D]. 北京. 中国科学院研究生院,2017.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
博士学位论文-文宏伟.pdf（5944KB）	学位论文		限制开放	CC BY-NC-SA