人脑MRI数据特征提取方法的研究与应用

项目名称： 人脑MRI数据特征提取方法的研究与应用

项目编号： No.61503411

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 苏龙飞

作者单位： 中国人民解放军军事科学院国防科技创新研究院

项目金额： 23万元

中文摘要： 建立功能磁共振成像（functional magnetic resonance imaging, fMRI）数据和人的神经、精神活动表现型的对应关系需要对fMRI数据进行精确的描述，这就是fMRI数据的特征提取。本课题拟利用机器学习的研究进展，从特征提取的角度对脑fMRI数据进行研究。由于在脑区间的相互作用中，线性关系约占95%，而另外5%为非线性相互作用。在对fMRI数据提取其线性功能连接特征的基础上，进一步提取其非线性功能连接特征来反映其非线性相互作用。在仿真数据上对提出的新特征进行测试后，将非线性特征应用于健康被试以及抑郁症和癫痫病人fMRI数据，可以进一步揭示人脑的发育以及抑郁症和癫痫的神经生理机制，为人脑发育以及脑疾病的网络化分析提供新视角。基于非线性功能连接特征的多变量模式分析对于解读人的情绪和思维具有重要意义，在临床上脑疾病的诊断也具有一定的意义。

中文关键词： 功能磁共振成像；特征提取；非线性功能连接；脑连接与脑网络；多变量模式分析

英文摘要： Feature extraction is essential for constructing the relationship between the functional magnetic resonance imaging (fMRI) data set and the corresponding experimental variables which reflect the human neural and mental activity. This program aimed at descripting the fMRI data set from the feature extraction point of view, which took advantage of the development in machine learning literature. It is believed that 95% of the various relationships between different brain regions are linear correlation, the other 5% were nonlinear relationships. Thus the linear functional connectivity and nonlinear connectivity features are both used for data description. On the simulated MRI data set, these novel feature extraction methods are tested. Then the new methods are applied to human fMRI data set of healthy subjects, subjects with major depression disorder and subjects with epilepsy, which will shed new light on the network analysis of brain development and the brain diseases from the functional integration perspective. The multivariate pattern analysis of fMRI data set based on these new measure will further explore the neural mechanism of the human mind and thought. Furthermore, this program has some potential in diagnosis and prevention of psychiatric disorders and neurodegenerative disease.

英文关键词： fMRI;feature extraction;nonlinear functional connectivity;brain connectivity and brain network;multivariate pattern analysis