【作者】 郭伟；

【导师】 顾实；

【作者基本信息】 电子科技大学 ， 工程硕士（专业学位）， 2021， 硕士

【摘要】 精神分裂症(Schizophrenia,SCZ)是一种慢性、持续的重大精神疾病,它在临床症状上表现各异,包括幻觉,妄想,行为异常,认知障碍和其他可变模式。精神分裂症病因病理至今尚未明确,分别有研究认为精神分裂症会影响大脑胼胝体结构和大脑功能交流。胼胝体(corpus callosum,CC)位于左右脑半球之间,是脑内主要的白质纤维束,起着整合两侧半球功能活动的重要作用。大脑功能连接是利用功能性核磁共振成像等非侵入式大脑成像技术将大脑功能交互作用绘制成相应的功能连接矩阵的技术,是研究大脑各区域功能交流及其强度的有力工具。尽管人们从各个方面对精神分裂症进行了研究,然而精神分裂症患者胼胝体特征与左右大脑半球间跨中线功能连接之间存在的关系仍然有待探索。本文应用一种稀疏典型相关分析(Sparse Canonical Correlation Analysis,SCCA)的方法,该方法能够计算出两组高维数据之间的相关关系,以此方法获得精神分裂症患者胼胝体特征与左右大脑半球间跨中线功能连接之间的相关性以及相关模式。本文首先从大脑功能连接中挑选出左右大脑之间的功能连接并应用双样本t检验对其进行特征选择以降维,然后标准化数据,之后利用网格搜索寻找设置SCCA所需要的两个超参数即作为正则稀疏化的两个1-范数惩罚因子,再使用最佳参数进行SCCA分析,最后使用置换检验来验证所得典型相关系数在统计学上是否有意义。作为补充,使用广义加性模型对所得功能连接做了年龄和性别效应的分析,以及使用支持向量机以左右大脑间功能连接作为输入特征建立分类模型。我们最后的结果是得到了五组典型向量,它们的相关系数依次为0.82、0.77、0.77、0.76、0.76,这些相关系数经过了显著性检验,其P值均小于0.05。稀疏典型相关分析给出的五组典型相关向量足以说明精神分裂症患者胼胝体FA值与组间有显著差异的左右大脑半球间跨中线功能连接的相关性比较高。而且,五组典型相关向量与之强相关的分别是胼胝体的五个部位,以及不同的左右大脑半球间跨中线功能连接。此外,精神分裂症患者胼胝体与功能连接之间的部分相关性和相关模式与患者年龄和性别有显著差异。最后我们得到的支持向量机分类模型准确率为0.8146。更多还原

【Abstract】 Schizophrenia(SCZ)is a chronic and persistent major mental illness with a variety of clinical symptoms,including hallucinations,delusions,behavioral abnormalities,cognitive disorders and other variable patterns.The etiology and pathology of schizophrenia are not yet clear.Some studies have suggested that schizophrenia affects the structure of the corpus callosum and the communication of brain function.The corpus callosum(CC),located between the left and right hemispheres,is the main white matter fiber bundle in the brain and plays an important role in integrating the functional activities of both hemispheres.Brain functional connection is a technology that uses non-invasive brain imaging techniques such as functional magnetic resonance imaging to draw the brain functional interaction into the corresponding functional connection matrix.it is a powerful tool to study the functional communication and intensity of various regions of the brain.Although schizophrenia has been studied from various aspects,the relationship between the characteristics of corpus callosum and cross-midline functional connection between left and right hemispheres remains to be explored.In this thesis,we use a sparse canonical correlation analysis(SCCA)method,which can calculate the correlation between two groups of high-dimensional data,and use this method to obtain the correlation between the characteristics of the corpus callosum and the cross-midline functional connection between the left and right hemispheres and the correlation patterns.In this thesis,we first select the functional connection between the left and right brain from the brain functional connection and use the double sample t-test to select the feature to reduce the dimension,then standardize the data,and then use grid search to find the two hyperparameters needed to set SCCA,that is,as two 1-norm penalty factors for regularization,and then use the best parameters for SCCA analysis.Finally,the Permutation test is used to verify whether the canonical correlation coefficient is statistically significant.As a supplement,the generalized additive model is used to analyze the age and gender effects of the functional connections,and the support vector machine is used to establish a classification model with the functional connections between the left and right brain as input features.Our final result is that five groups of canonical vectors are obtained,and their correlation coefficients are 0.82,0.77,0.77,0.76 and 0.76 respectively.These correlation coefficients have been tested for significance,and their P values are all less than 0.05.The five groups of canonical correlation vectors given by sparse canonical correlation analysis show that there is a high correlation between the FA value of corpus callosum and the cross-midline functional connection between the left and right hemispheres in patients with schizophrenia.Moreover,the five canonical correlation vectors are strongly correlated with the five parts of the corpus callosum and the cross-midline functional connections between the left and right hemispheres.In addition,the partial correlation and correlation pattern between corpus callosum and functional connection in patients with schizophrenia were significantly different from those of patients’ age and sex.Finally,the accuracy of the support vector machine classification model is 0.8146.更多还原