CASIA OpenIR  > 毕业生  > 博士学位论文
基于磁共振影像的脑结构形态学研究与应用
其他题名Morphological Analysis of Brain Structures Based on Magnetic Resonance Imaging, and its Applications
石峰
学位类型工学博士
导师蒋田仔
2008-05-29
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业模式识别与智能系统
关键词计算神经解剖学 磁共振影像 海马 Meta分析 Vbm 皮层厚度 Computational Neuroanatomy Magnetic Resonance Imaging Hippocampus Meta Analysis Vbm Cortical Thickness
摘要传统的磁共振影像学分析主要关注感兴趣区域的体积以及在全脑范围内寻找体积有差异的区域。随着计算机视觉和数学领域相关知识的引入,图像分割与配准方法日渐成熟,研究热点逐步转移到感兴趣区域的形状建模以及形状差异分析上。本论文正是以磁共振影像学分析为主要内容,从传统体积测量分析方法的深化到新的形态学分析方法,都提出了一些解决方案,并取得了一定成果。主要贡献如下: 1. 针对在阿尔茨海默病和轻度认知障碍人群中,不同的研究报道的海马体积变化相互矛盾,而且极少有研究关注海马的左右对称性变化的现状,我们引入文献汇总分析方法——Meta方法,通过分析28篇代表性文献中海马体积变化的趋势以及程度,整合得出海马在两种疾病中的萎缩程度,并指出海马存在着左小于右的不对称性,而且这种不对称性在轻度认知障碍人群中最大,在阿尔茨海默病中最小。本研究结果提示在疾病发展的初期,海马体积就会发生变化,而且除了体积以外,本文发现的不对称性也可以作为潜在的生物标记,用来对疾病的发生以及程度进行预测。 2. 针对海马形状分析的难题,我们把香蕉形状模型引入到海马形状建模中,用400×200的表面网格表示海马曲面,并提出纵向曲线配准方法进一步优化对应性,使用表面顶点到中轴的径向距离作为海马膨胀与萎缩的度量,并应用到19名阿尔茨海默病患者和20名正常对照的研究中,发现左海马显著萎缩,主要集中在海马外侧和中部,而右海马没有明显的萎缩区域。我们的形状分析方法不仅能得到与体积测量一致的结果,而且可以更精细的定位海马的异常部位,有利于研究异常萎缩的病理学原因。 3. 针对盲人由于视觉剥夺导致的皮层发育异常的问题,我们首先用基于体素的全脑形态学方法(VBM)在全脑范围寻找异常区域。发现枕叶视交叉区域双侧白质显著萎缩,枕极楔叶灰质萎缩而附近的左侧舌回却有异常膨胀。针对枕极灰质的异常情况,采用皮层建模的方法进一步分析皮层厚度差异。厚度分析发现,盲人双侧枕叶具有显著较高的皮层厚度,然而显著区域的面积则比正常人小。这表明盲人皮层水平方向上的面积萎缩和垂直方向上的厚度膨胀是共存的。我们的发现提示,由于没有视觉经验的后天刺激,盲人在发育过程中修剪过程和髓鞘化过程可能被阻断而导致皮层发育异常。 4. 针对影像学特征与临床诊断的结合,我们提出一种特征选择方法,并应用于精神分裂症病人和正常人的自动分类。首先从功能磁共振图像的时间序列中提取局部一致性特征,经过特征选择和降维过程,用线性分类器对测试图像分类,并在交叉验证的条件下达到了80%的分类正确率。结果表明用影像学特征实现对疾病的自动分类是可行的。另外通过跟踪权重最大的区域,还可以反过来定位哪些区域对疾病最敏感。
其他摘要Traditional magnetic resonance imaging analysis methods measure the volume of region of interests (ROI), as well as the regional difference in whole brain. As the introduction of many computer vision and mathematics methods, the segmentation and registration methods are widely used and highly improved, and the surface reconstruction and shape analysis are getting more attention. This paper proposes several methods on both traditional volume counting and shape analysis. The main contributions are as follows: 1. In patients of Alzheimer’s Disease (AD) and Mild Cognitive Impairment (MCI), the studies on hippocampus volume report controversial results, and few of them concerns the left-right asymmetry of hippocampus. We introduce Meta-analysis method to integrate the results. By analyzing hippocampus volume differnece in 28 papers, the combined atrophy rates are obtained in AD and MCI. The findings also show a left-less-than-right pattern, which is maximum in MCI and minimum in AD. Our results find the hippocampus is involved in AD and MCI early, and the asymmery could be a candidate biomarker besides hippocampus volume. 2. We introduce the Banana-like model in hippocampus surface reconstruction, and describe its shape with 400*200 grids. The area of ISO-circle is registered to the mean area curve and point correspondence is improved in this way. The radial distance from each vertex on surface to the mid-axis is taken as measurement. After applying to 19 Alzheimer’s Disease patients and 20 normal controls, significant atrophy is found in left hippocampus, mainly in lateral and middle regions. This findings are consistant with the volume counting results but provide more precise location of abnormal regions, which could help to study the pathology reasons of hippocampus atrophy. 3. We employ VBM method to find abnormal regions of whole brain in early blind, who may experience abnormal cortical development process caused by visual deprivation. Results show significant bilateral white matter and cuneus atrophy, but left lingual gyrus is inflated. Then, a cortex modeling method is used. The findings show early blindness have higher cortical thickness in Brodmann 17/18, but the area of that region is smaller than that of controls. This reveals the horizontal shrinking is concurrent with vertical inflation. This finding suggests the synaptic pruning and myelination process may be interrupted by lacking of visual experience during brain developing. 4. We propose a pattern classification method for combing the neuroimaging properties and clinical diagnosis. Features are first extracted from regional homogeneity attribute in time-series of fMRI. After feature selection and dimension reduction, we get the discriminative model with training images and achive 80% average correct rate in leave-one-out cross validation. This suggests the automatic patients classification based on neuroimaging measurements is feasible.
馆藏号XWLW1197
其他标识符200418014628019
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6086
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
石峰. 基于磁共振影像的脑结构形态学研究与应用[D]. 中国科学院自动化研究所. 中国科学院研究生院,2008.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20041801462801(3485KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[石峰]的文章
百度学术
百度学术中相似的文章
[石峰]的文章
必应学术
必应学术中相似的文章
[石峰]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。