摘要
为有效提取轻度认知障碍患者发生病变的特征、解决传统聚类方法在BOLD-fMRI数据特征提取中存在的不足,提出了改进的谱聚类方法。采用改进的谱聚类方法对fMRI数据进行聚类、提取模式特征,使用该特征对MCI和正常人进行分类研究,正确率达到了82%,且存在异常模式的脑区大多属于MCI患者的关键脑区。实验结果表明,谱聚类可以应用于BOLD-fMRI特征提取,为今后MCI辅助诊断模型提供了一定的研究基础。
To extract the appropriate features from patients of magnetic resonance imaging and solve the deficiencies of traditional clustering methods on feature extraction of the BOLD-fMRI data, an improved spectral clustering method is put forward. An im- proved spectral clustering technique is used to cluster the fMRI data, extract feature mode, and classify subjects of MCI and nor- mal, the correct rate reaches 820/oo. Furthermore, the abnormal patterns of brain areas mostly exist in key brain areas of MCI. The results show that spectral clustering can be applied to BOLD-fMRI feature extraction, which provides a foundation for future research of MCI diagnosis model.
出处
《计算机工程与设计》
CSCD
北大核心
2014年第4期1379-1384,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61070077
61170136)
山西省自然科学基金项目(2010011020-2
2011011015-4)
北京市博士后工作经费基金项目(Q6002020201201)
关键词
轻度认知障碍
谱聚类
BOLD变化率
体素
SVM分类
mild cognitive impairment
spectral clustering
BOLD response rate
voxels
SVM classification
