机构地区: 第四军医大学唐都医院
出 处: 《西安工业大学学报》 2012年第8期661-664,共4页
摘 要: 大脑结构MRI数据本质上是三维张量数据,而传统机器学习方法在提取大脑结构特性信息时,需要将其展开为一维向量,破坏了数据的内在结构信息.为了克服数据向量化带来的缺点,提出使用张量线性判别分析算法,用于大脑结构MRI数据分析.并对比基于向量的主成分线性判别分析算法,对20个正常人和20个精神病患者的脑灰质MRI结像进行分类.结果表明张量线性判别分析算法的最高识别率达到95%,其总体识别率、鲁棒性都要优于主成分线性判别分析算法.张量线性判别分析算法在大脑MRI数据分类上要优于传统基于向量的机器学习方法. Essentially, brain's MRI data are three-dimensional tensor data. Traditional vector-based machine learning methods typically represent MRI data as one-dimensional vectors when extracting information of the brain structure,which breaks the natural structure of the original data, so that some potentially useful information underlying the neuroimaging data is ignored. To overcome these drawbacks , tensor LDA method is put forward to analyze brain's MRI data. MRI data of 20 normal persons and 20 schizophrenic patients were analyzed by Tensor LDA method contrast to vector-based PCA +LDA method. The result showed that the overall recognition rate, top recognition rate and robustness of Tensor LDA method are superior to PCA+LDA method. Tensor LDA method outperforms the vector- based methods.
关 键 词: 磁共振成像 张量线性判别分析 大脑结构像 精神分裂症
领 域: [自动化与计算机技术] [自动化与计算机技术]