静息态脑功能网络分析的假设驱动和数据驱动方法综述

Review of hypothesis driven and data driven methods for brain functional network analysis

基于功能磁共振成像的研究已发现在脑区之间存在低频波动一致的脑功能网络。脑功能网络有助于认识脑功能和诊断神经精神疾病,而脑功能网络分析方法在其中具有重要地位。本文首先综述了假设驱动和数据驱动进行脑功能网络分析的两类方法,接着对常用的基于感兴趣区、基于体素、基于独立成分分析、基于主成分分析和基于聚类方法的原理、优缺点做了详细介绍,并着重阐述了最近提出的组信息指导的独立成分分析方法,及基于半监督学习技术的感兴趣区选择方法,最后对改进方向进行了展望。

The research of functional magnetic resonance imaging （fMRI） has revealed the coherent fluctuation of low frequency signals between anatomically separated brain regions, which indicates functional network. Brain functional networks help better understanding of brain function and diagnosis of mental disease, and methods for analyzing brain functional networks play an important role. The paper firstly reviews two categories of methods for functional networks analysis： hypothesis driven and data driven methods, and then presents the theory, advantages and shortcomings of the traditional methods,including region of interests （ROI） based method, voxel based method, independent component analysis （ICA）, principal component analysis （PCA） and clustering method. Furthermore, the paper highlights the recently proposed group information guided ICA method and ROI selection method semi-supervised learning based Finally, future potential improvements of the methods are prospected .