功能磁共振成像无监督模式分析方法及应用

项目名称： 功能磁共振成像无监督模式分析方法及应用

项目编号： No.61503397

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： 曾令李

作者单位： 中国人民解放军国防科技大学

项目金额： 22万元

中文摘要： 如何从功能磁共振成像(fMRI)中提取感兴趣的脑活动模式，是脑影像分析方法学研究最具有挑战性的课题之一。多变量模式分析方法可以同时利用多个变量来分析组间差异，比单变量统计学方法表现出更好的性能，而无监督模式分析方法可以在没有类别信息的情况下非常灵活地确定数据的复杂未知分布或模式。但是，目前fMRI无监督模式分析方法研究较少。本项目将开展fMRI无监督模式分析理论与方法研究，重点研究fMRI二维皮层表面特征描述、多流形学习等机器学习方法来提取fMRI数据中的低维本征模式，研究多流形聚类和最大间隔聚类等算法来解决低信噪比、数据复杂分布的fMRI数据聚类问题，并围绕精神类疾病亚型分析的临床课题开展应用研究。期望建立fMRI无监督模式分析的一般性方法框架，找到精神分裂症和重度抑郁症脑网络机制的联系和区别，以及支持精神分裂症若干分型的影像学证据和可以预测精神分裂症患者药物治疗响应的潜在影像学标记。

中文关键词： 功能磁共振成像；无监督学习；多流形聚类；脑连接；亚型分析

英文摘要： To extract brain activity patterns of interest from functional MRI data is one of the biggest challenges in the methodology of brain imaging analysis. Multivariate pattern analysis can take advantage of multiple classification features simultaneously to analyze group differences, so it always performs better than univariate statistical analysis. Unsupervised pattern analysis approaches can discover homogeneous groups (called clusters) of similar samples within the data and flexibly determine the complex and unknown distribution or pattern of data within the input space when a priori labeling information is unavailable. To date, however, little attention has been paid to unsupervised pattern analysis methods of neuroimaging data. This research proposal will study the theories and methods of unsupervised pattern analysis, focusing on two-dimensional feature descriptors of fMRI-based cortical surface and multi-manifold learning for exploring low-dimensional intrinsic brain activity patterns from fMRI data, as well as multi-manifold clustering and maximal margin clustering algorithms for solving the clustering of high-dimensional fMRI data with low signal-to-noise ratio and complex distribution. Using the developed methods, we attempt to perform subtype analysis of psychiatric disorders. We expect to develop a general framework of unsupervised pattern analysis of fMRI data and discover the convergence and divergence of brain network mechanisms between schizophrenia and major depression, and neuroimaging-based evidence of some schizophrenic subtypes, as well as potential neuroimaging-based markers to predict outcome of medication in schizophrenia.

英文关键词： fMRI;unsupervised learning;multi-manifold clustering;brain connectivity;subtype analysis