基于曲面形态学的轻度认知障碍脑结构特性研究

发布时间：2018-09-12 16:16

【摘要】：大脑是人类最重要也是最复杂的器官,深入研究大脑的结构特性和工作机制对于保护大脑的健康、更好的开发大脑的潜能具有深远的意义。轻度认知障碍(MCI)被认为是介于正常老年人(NC)和阿尔茨海默症(AD)之间的过渡形态,具有认知损害临床表现不明显、病情不稳定等特点。由于AD具有病情不可逆性,因此研究MCI脑结构的变化机理对于降低AD的发病率和死亡率具有重要意义。基于曲面的形态学是近年来新出现的一种大脑形态学分析方法,被越来越多的研究证明在大脑皮层(尤其在沟回区)的测定上具有独特的精确度和灵敏度。目前对于MCI的曲面形态学研究大多集中在MCI与其它群体皮层厚度的横向比较上,其他方面的研究还比较少。因此本文结合了磁共振图像数据和基于曲面的形态学方法对MCI进行了多方面的研究,希望借此加深对于MCI发病机制及其对大脑结构影响的了解,更好的实现对于老年痴呆的早期预防和早期治疗。本文的主要贡献如下： 1.针对目前对于MCI群体脑结构异常尚无多时间点纵向跟踪研究的情况,首次提出利用基于曲面的大脑形态学方法对于MCI大脑萎缩模式进行纵向跟踪分析。在对于MCI和NC群体的结构磁共振图像数字化处理之后,得到了大脑皮层的曲面形态学指标。然后对MCI和NC群体在两年时间跨度上的大脑皮层厚度进行了比较,分析了皮层厚度萎缩速度的差异以及厚度萎缩与简易智能状态检查量表(MMSE)得分之间的相关性。更进一步的我们分析了MCI中的异常脑区在更长时间跨度上(四年)脑区平均厚度的变化趋势。结果表明MCI群体相对NC在某些脑区如颞叶、脑岛、海马旁回上存在显著性的厚度变薄异常,这些异常脑区的平均厚度随时间呈现显著性的线性下降趋势且速率大于NC群体,而且相关性结果表明这些脑区的萎缩与临床表现的下降有直接的相关性。 2.由于目前尚无使用皮层厚度构建MCI结构脑网络的研究,我们首次使用了脑区平均厚度构建了MCI与NC的结构脑网络,并且发现了两组群体在小世界属性上的显著差异。首先使用自动解剖分析模板(AAL)进行了脑区的划分,计算出脑区的平均厚度值,然后基于脑区间厚度的偏相关系数矩阵构建了结构脑网络。我们使用了置换检验、Fisher Z变换等操作比较了两组人在平均聚类系数、平均最短路径长度、hub节点集中度以及脑区间连接相关性上的区别。结果表明MCI群体相对于NC在各个稀疏度下均表现出更大的聚类系数和更大的最短路径长度的趋势,并且在某些稀疏度下差异具有显著性。同时在MCI脑网络中也存在着增加和缺失hub节点以及增加和减少脑区间连接性的情况。该研究结果与他人使用功能磁共振数据或者灰质体积构建MCI脑网络的研究结果较为吻合,进一步说明了MCI确实从某种程度上改变了大脑处理信息的机制。 3.为了将大脑形态学的研究与临床早期疾病预测更好的结合起来,同时也为了测试皮层厚度指标的可靠性和准确性,我们提出了一种基于皮层厚度的特征选择方法,并应用到了转化型轻度认知障碍(CMCI)和稳定型轻度认知障碍(SMCI)的分类上。首先使用统计分析方法验证了NC、CMCI和SMCI三组群体在基线时间(baseline)上确实存在着显著性的厚度差异,厚度趋势为NCSMCICMCI,为接下来利用厚度数据进行模式分类提供了依据。然后计算出了两组人78个脑区的平均厚度值,并将其作为模式分类的特征向量。我们比较了两种特征选择方法对于分类结果的影响,一种是根据脑区厚度差异的显著性来作为特征选择的依据,一种是利用联合递归特征去除(RFE)算法得到的特征排序系数作为特征的选择依据。在分类过程中选择了径向基核函数(RBF)作为支持向量机(SVM)的核函数,并进行了网格参数寻优。分类结果表明利用脑区平均厚度来进行轻度认知障碍的转化预测存在一定的可行性,分类准确率在留一交叉验证下最高达到76.77%。
[Abstract]:The brain is the most important and complex organ in human beings. It is of profound significance to study the structural characteristics and working mechanism of the brain to protect the health of the brain and better develop the potential of the brain. Mild cognitive impairment (MCI) is considered to be a transitional form between normal elderly (NC) and Alzheimer's disease (AD) with cognitive impairment. Because AD is irreversible, it is important to study the mechanism of brain structure changes in MCI to reduce the incidence and mortality of AD.
Surface-based morphology is a new method of brain morphology in recent years. It has been proved by more and more studies that it has a unique accuracy and sensitivity in the measurement of cerebral cortex (especially in the sulcus area). Therefore, this paper combines magnetic resonance image data and surface-based morphology to study MCI in many aspects, hoping to deepen the understanding of the pathogenesis of MCI and its impact on brain structure, and better realize the early prevention and treatment of Alzheimer's disease. The contributions are as follows:
1. In view of the fact that there are not many longitudinal tracking studies on brain structural abnormalities in MCI population at present, the longitudinal tracking analysis of brain atrophy patterns in MCI population is first proposed by using surface-based brain morphology. After digitizing the structural magnetic resonance images of MCI and NC population, the curved surface of cerebral cortex is obtained. Then we compared the cortical thickness of the MCI and NC populations over a two-year period, analyzed the difference in the rate of cortical thickness atrophy and the correlation between the thickness atrophy and the score of the Simple Intelligent State Examination (MMSE). The results showed that there were significant thinning abnormalities in MCI population relative to NC in some brain areas such as temporal lobe, insula and parahippocampal gyrus. The average thickness of these abnormal brain areas showed a significant linear downward trend with time and the rate was higher than that of NC population, and the correlation results showed that these abnormal brain areas were thinner than NC population. Atrophy is directly related to the decline of clinical manifestations.
2. Since there is no research on using cortical thickness to construct MCI brain network, we constructed MCI and NC brain network using the average thickness of the brain area for the first time, and found significant differences in small-world attributes between the two groups. Then we constructed the structured brain network based on the partial correlation coefficient matrix of brain interval thickness. We compared the differences between the two groups in average clustering coefficient, average shortest path length, hub node concentration and the correlation of brain interval connectivity using permutation test and Fisher Z transform. The clustering coefficients and shortest path lengths were larger in all sparsities, and the differences were significant in some sparsities. At the same time, there were also cases of increasing and missing hub nodes and increasing and decreasing brain region connectivity in MCI brain networks. The results of data or gray matter volume building MCI brain networks are more consistent, further demonstrating that MCI does, to some extent, alter the mechanism by which the brain processes information.
3. In order to combine the study of brain morphology with the prediction of early clinical diseases, and to test the reliability and accuracy of cortical thickness, we propose a feature selection method based on cortical thickness, which is applied to the classification of transformed mild cognitive impairment (CMCI) and stable mild cognitive impairment (SMCI). Firstly, the statistical analysis method was used to verify the significant difference of thickness between NC, CMCI and SMCI groups at baseline time. Thickness trend was NCSMCICMCI, which provided a basis for pattern classification using thicknesses data. Then the average thicknesses of 78 brain regions were calculated and compared between the two groups. For feature vectors of pattern classification, we compare the effects of two feature selection methods on classification results. One is based on the significance of the difference in brain area thickness, the other is based on the feature ranking coefficient obtained by the joint recursive feature removal (RFE) algorithm. Radial Basis Function (RBF) was selected as the kernel function of Support Vector Machine (SVM) and the grid parameters were optimized. The classification results showed that it was feasible to predict the transformation of mild cognitive impairment by using the average brain area thickness. The classification accuracy was up to 76.77% under the left-one cross validation.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R445.2;R749.1

【共引文献】