CASIA OpenIR  > 毕业生  > 硕士学位论文
磁共振神经影像数据分类方法的研究及其应用
Alternative TitleResearch on the classification methods of magnatic resonance neuroimaging data and its application
戴岱
Subtype工学硕士
Thesis Advisor何晖光
2012-05-29
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword神经影像 模式分类 机器学习 阿尔茨海默病 注意力缺失/过动症 Neuroimaging Pattern Classification Machine Learning Alzheimer's Disease Attention-deficit/hyperactivity Disorder
Abstract随着人口老龄化进程的加快以及人们承受精神压力的增大,神经与精神疾病的发病率在世界范围内逐年攀升,已引起了越来越多的医生和研究人员的注意。神经影像技术尤其是磁共振技术的发展使得人们可以“无创”的观察大脑的结构和功能,为脑部疾病的诊断及大脑工作原理的研究提供了新的手段。但是,由于人脑复杂的结构以及日益增长的巨大数据量,医生和研究人员难以高效的从神经影像数据集中提取出疾病相关的信息并作出正确的诊断。近年来,模式识别技术在神经影像领域中的应用使得脑部疾病的计算机辅助诊断和病灶定位成为了可能。本文介绍了当前神经影像模式分类的最新研究现况,并提出了预测阿尔茨海默病人和注意力缺失/过动症儿童的分类方法,主要的工作体现在如下几个方面: (1)针对阿尔茨海默病的分类,本文提出了一个基于脑皮层厚度网络的分类方法。与常规的构建组间网络的方法不同,这种方法使用不同区域的平均皮层厚度为每个被试构建个体网络,并将网络边的权重作为特征训练分类器,其在OASIS公开数据集上的分类结果为90.4%,达到甚至超过国内外最新研究结果。另外,在该方法中还使用了一种混合的特征选择方法,保证特征选择精度的同时可以减小计算复杂度。 (2)对于注意力缺失/过动症的分类,本文提出了一个融合多模态影像特征的方法,首先从磁共振结构图像和静息态功能图像中分别提取皮层厚度、灰质密度、局部一致性和功能连接网络等特征,然后为每个特征指定不同的核函数或核参数,最后训练多核学习分类器为每个核函数分配权重,融合全部特征。在ADHD-200公开数据集上的实验中发现多核学习融合多模态特征的分类性能优于使用单一特征的情况。我们使用该研究的方法参加了注意力缺失/过动症国际分类竞赛,在众多国际知名大学和科研机构的队伍中取得了两项指标第一、总分第六的优秀成绩。(3)为了辅助脑网络研究,我们开发了脑网络分析与可视化工具包VisualConnectome。该工具包可以方便的对脑网络进行可视化并计算网络拓扑属性,使研究流程变得简洁而直观,极大的方便了相关研究和论文工作。
Other AbstractWith the growth of aging population and the heavier burden that people bear in the modern world, the incidence of neurological and mental disorders grows year by year worldwide, which draws the attention of more and more medical doctors and neuroimaging researchers. The development of neuroimaging technologies, especially magnetic resonance imaging (MRI) technology, makes it possible for people to observe the brain structures and functions non invasively, providing a new avenue for diagnosis of brain disease and researches on brain functions. However, due to the complexity of brain structure and the continuously increased amount of imaging data, it is difficult for researchers and medical doctors to efficiently extract helpful information that is relevant to the disease, or to make correct diagnosis. In recent years, the application of pattern recognition technologies in neuroimaging community makes possible the computer aided diagnosis and localization of disease focus. This paper introduces the recent researches related to pattern classification in neuroimaging community, and puts forward the classification frameworks for prediction of Alzheimer's patients and children with attention deficit/hyperactivity disorder. The main contributions of this paper are as the following: (1) For classification of Alzheimer's disease, this paper presents a classification framework based on cortical thickness networks. In this work, we construct a network for each subject using the mean cortical thickness of different brain regions, and the edge weight of these networks are used as features to train an SVM classifier. Because the network features pay more attention on the relationships of cortical thickness change between different brain areas which is caused by the diseases, they are not sensitive to the individual differences and therefore are more stable and reliable. The experiments on the public OASIS datasets show that the best accurary of our method is 90.4%, reaching or even exceeding the results of related studies at home and abroad. In addition, we also use a hybrid method for feature selection, which can ensure the performance and reduce the computational complexity of the feature selection process. (2) As for classification of attention deficit/hyperactivity disorder, this paper presents a classification framework based on multi-kernel learning to fuse the features extracted from multimodal imaging. First several features such as cortical thickness, gray...
shelfnumXWLW1775
Other Identifier200928014628029
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7635
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
戴岱. 磁共振神经影像数据分类方法的研究及其应用[D]. 中国科学院自动化研究所. 中国科学院研究生院,2012.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20092801462802(7078KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[戴岱]'s Articles
Baidu academic
Similar articles in Baidu academic
[戴岱]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[戴岱]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.