多模态fMRI信息融合的脑功能网络构建与分析

As the improvement of magnetic resonance imaging quality and the greatest development on latest research, researchers can accurately locate and define different brain areas. In recent years, brain functional connectivity has become the hot spots of brain cognitive research, mainly through the analysis of brain cognitive data obtained in the single state of the task state or resting state. The proceeds of the brain network results or have strong task dependence or lack of access to cognitive explanation.The calculated brain networks are depends on the task that have been used greatly or are very difficutly explained in brain cognition. Therefore,the idea of multi-modal combination is proposed in this project.That is, the data from tasking state and resting state is used at the same time. based on the characters of clarity cognition in tasking and function default in resting, we investigate the model of brain network and cognition computing.The main contents include: a rational and efficient dimensionality reduction study on mass brain image data based on wavelet compressive sensing space; The study of effective brain activation characteristics based on the frequency domain of multi-modal fMRI; Modeling and analysis of resting state brain network based on Bayesian networks. The study of semi-supervised brain network model based on multi-modal fMRI; The study of cognition description of classic brain network models and the cognitive characteristics of different functional magnetic resonance imaging data. The study of relationships of differnet cognitive models and cognitive data and based on which exploring the basic mechanisms of cognitive computing. This project comprehensively investigates the construction of brain network and deeply explores brain cognition.

随着核磁共振成像质量的不断提高和最新研究的跨越式进展，研究人员可精确地定位与细分不同脑区的功能。近年来，脑功能性连接已成为脑认知研究的热点，主要通过分析任务态或静息态等单一状态下所获取的脑认知数据来完成，所得脑网络结果或者具有较强的任务依赖，或者难以获得认知解释。因此，本课题提出任务与静息多模态下脑功能网络的研究思路，利用各任务态下认知功能明确、静息态下功能默认的特点，综合研究脑网络模型，探讨认知计算。主要研究内容包括：基于小波压缩传感空间的海量脑影像数据合理高效降维研究；基于频率的多模态fMRI中有效脑激活特征研究；基于贝叶斯网络的静息态脑网络建模与分析；基于多模态fMRI的半监督脑网络模型研究;开展各脑网络模型认知表达与各模态功能核磁共振成像数据认知特点的研究；在上述认知模型与认知数据关系研究的基础上，探索认知计算的基本机理。本课题的开展，既全面研究脑网络构建，还深度理解脑认知。

随着核磁共振成像技术的提高，依托其成像数据进行脑认知状态判别与分析研究已然成为研究热点。然而，数据的高维、高冗余、强状态依赖是研究人员必须克服的挑战。为此，本项目面向脑认知影像大数据，特别是核磁共振影像数据，围绕脑认知状态、活动等判别分析中的关键科学问题开展研究。提出了约束的稀疏表示特征计算模型、改进了模糊C均值分割算法，提高了大变形脑图象的定位与分割精度，构建了基于多变量Bayesian网络的脑网络分析模型、基于深度神经网络ADHD fMRI脑网络分析模型、基于社会网络的ADHD fMRI脑网络分析模型、基于卷积神经网络ADHD fMRI脑网络分析模型等一系列脑网络分析模型，并研究了不同脑区的认知功能及其对ADHD的判别分析影响。上述成果为基于大数据的脑认知研究提供了理论方法，为将来的数字医疗提供技术支持。. 本项目在《IEEE Transactions on Systems，Man, and Cybernetics: Systems》、《IEEE Transactions on Wireless Communications》、《Neurocomputing》、《Digital Signal Processing》、《Bio-Medical Materials and Engineering》等国内外学术刊物和会议上共发表论文（含录用）19篇，事实资助但未标注论文（含录用）4篇（任何基金未标注），其中，SCI检索10篇。申请并公开发明专利3项，软件著作权7项。培养1名博士生，8名硕士生。项目负责人何良华教授获得2014年度教育部科技进步一等奖1项（排名第9），并入选第一届教育部“长江学者奖励计划”青年学者项目。