基于稠密子图挖掘的脑网络分类研究及应用

发布时间：2018-07-12 11:53

本文选题：脑网络分类 + 稠密子图分割　；参考：《南京航空航天大学》2017年硕士论文

【摘要】：脑疾病不仅威胁到患者的生命,而且为社会带来沉重的负担,因此人们越来越关注脑疾病的研究。随着神经影像技术的快速发展,研究人员将神经影像技术应用在脑疾病研究上,并且取得了极大的进展。研究人员通过从神经影像中挖掘出脑功能连接网络,然后应用机器学习和数据挖掘的方法从中挖掘特征,将其应用于脑网络分类。本文的研究基于脑网络研究脑疾病,主要工作如下:第一部分,我们提出了基于稠密子图分割的脑网络分类方法。此部分的工作主要利用稠密子图来反映人类大脑中连接紧密的拓扑模式,具体而言,通过将脑网络转化为边对偶图,将转化后的对偶图分割为稠密度依次降低的稠密子图,然后挑选出稠密度最大的若干稠密子图,使用挑选出的稠密子图重构原有的脑网络。最后使用图核度量重构后的脑网络之间的相似性,构建图核矩阵,使用SVM进行分类。实验结果表明,提出的分类方法可以提升分类效果,并且可以挖掘出与脑疾病相关联的区域。第二部分,我们在稠密子图的基础上结合频繁子图的概念,提出了基于频繁稠密子图的脑网络分类方法。此部分的工作主要基于脑疾病患者和正常人之间的拓扑结构出现显著性差异的事实,因此从患者数据集和正常人数据集中挖掘出的频繁稠密子图会存在显著差异,可以用于脑网络分类。具体而言:首先分别从患者和正常人数据集中挖掘出频繁稠密子图,然后计算每一个子图的判别性,挑选出最具有判别性的稠密子图并且构建指示矩阵作为特征矩阵,最后使用SVM分类。实验结果表明,基于频繁稠密子图的分类方法不仅可以提高分类的精度,而且可以挖掘出与脑疾病相关的拓扑模式。
[Abstract]:Brain diseases not only threaten the lives of patients, but also bring a heavy burden to the society, so people pay more and more attention to the research of brain diseases. With the rapid development of neuroimaging technology, researchers applied neuroimaging technology to brain disease research, and made great progress. The researchers used neural images to extract brain functional connectivity networks and then used machine learning and data mining methods to extract features from them and apply them to the classification of brain networks. The main work of this paper is as follows: in the first part, we propose a classification method of brain network based on dense subgraph segmentation. In this part, dense subgraphs are mainly used to reflect the closely connected topological patterns in the human brain. In particular, by transforming the brain network into edge-pair bigraphs, the transformed digraphs are divided into dense subgraphs with decreasing density in turn. Then some dense subgraphs with the highest density are selected and the original brain network is reconstructed using the selected dense subgraphs. Finally, the similarity of the reconstructed brain network is measured by using the graph kernel, and the kernel matrix is constructed, and SVM is used to classify the reconstructed brain network. The experimental results show that the proposed classification method can improve the classification effect and can mine the regions associated with brain diseases. In the second part, combining the concept of frequent subgraph with dense subgraph, we propose a classification method of brain network based on frequent dense subgraph. This part of the work is mainly based on the fact that there are significant differences in topological structure between patients with brain disease and normal people, so there are significant differences between frequent dense subgraphs mined from patient data sets and normal person data sets. Can be used for brain network classification. Specifically, the frequent dense subgraphs are mined from the patient and normal data sets respectively, and then the discriminant properties of each subgraph are calculated, the dense subgraphs with the most discriminant characteristics are selected, and the indicator matrix is constructed as the characteristic matrix. Finally, SVM classification is used. The experimental results show that the classification method based on frequent dense subgraphs can not only improve the classification accuracy but also mine the topological patterns related to brain diseases.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R741.044;TP391.41

【相似文献】