基于大脑结构网络特征的遗忘型轻度认知障碍诊断模型

发布时间：2019-06-18 14:03

【摘要】：选取与认知表现分数相关性高的结构网络拓扑特征,利用这些特征建立分类模型,实现对正常老化者及遗忘型轻度认知障碍(aMCI)患者的分类。本研究包含两组扩散张量影像(DTI)数据,一组为52例正常老化受试者,一组为39例aMCI患者。分别对两组数据进行结构网络构建,采用图论分析法提取结构网络的特征,将所有特征与简单智能状态检查量表(MMSE)分数进行相关性分析,选取与认知表现分数高度相关的特征,基于这些特征建立5种分类模型,并对模型的分类效果进行评估。对于正常老化数据,选出18个与认知能力显著相关的结构网络特征,集中于解剖自动贴标(AAL)图谱中的9个脑区;对于aMCI数据,也选出18个与认知能力显著相关的结构网络特征,集中于AAL图谱中的9个脑区;二者选出的特征及分布的脑区是不同的。通过对分类模型的评估,得出支持向量机序列最小优化算法建立的模型分类效果最佳,特异性达到88.46%,敏感性达到83.05%,准确性达到85.71%。所提取的与认知表现相关性高的结构网络特征,可以作为生物标记指针,来建立分类模型,对正常老化者及aMCI患者进行分类,也可提供相应脑区间连接变化的信息。
[Abstract]:The structural network topological features with high correlation with cognitive performance scores were selected and used to establish classification models to classify (aMCI) patients with normal aging and amnesia mild cognitive impairment. This study included two groups of diffusion tensor imaging (DTI) data, one group of 52 normal aging subjects and the other group of 39 patients with aMCI. The structural network of the two groups of data is constructed respectively. The features of the structural network are extracted by graph theory analysis. The correlation between all the features and the (MMSE) score of the simple Intelligent State examination scale is analyzed. The features highly related to the cognitive performance score are selected. Based on these features, five classification models are established, and the classification effect of the model is evaluated. For the normal aging data, 18 structural network features significantly related to cognitive ability were selected, focusing on 9 brain regions in the anatomical automatic labeling (AAL) map, and for aMCI data, 18 structural network features significantly related to cognitive ability were also selected, concentrated in 9 brain regions in the AAL map, and the selected characteristics and distributed brain regions were different. Through the evaluation of the classification model, it is concluded that the model established by the support vector machine sequence minimum optimization algorithm has the best classification effect, the specificity is 88.46%, the sensitivity is 83.05%, and the accuracy is 85.71%. The extracted structural network features, which are highly correlated with cognitive performance, can be used as biomarkers to establish classification models to classify normal aging patients and aMCI patients, and can also provide information on the changes of connections between brain regions.
【作者单位】：北京工业大学生命科学与生物工程学院;长庚大学资讯工程学系;长庚大学健康老化中心;阳明大学生物医学影像暨放射科学系;阳明大学医学系;
【基金】：长庚大学研究计划(UERPD2B0301,UERPD2C0041) 长庚大学健康老化中心计划(CMRPD1B0331,EMRPD1D0261)
【分类号】：R749.1

【相似文献】