医学图像配准在肝脏DCE-MRI和静息态fMRI中的应用

发布时间：2018-02-25 03:30

本文关键词： 图像配准肝脏DCE-MRI 静息态fMRI 低秩分解流形学习功能相关性张量自由形变模型 LDDMM配准　出处：《南方医科大学》2017年博士论文　论文类型：学位论文

【摘要】：医学图像在临床上的应用,比如疾病监测和诊断,愈合结果的评判等等,使其研究和分析迅速成为一个新兴领域。随着图像尺寸的增加,医学图像可以提供更多更丰富的医学信息,这些庞大的信息需要及时和准确的解释和分析。因此,医学图像配准就变得十分重要且必要,这种处理方式可以让不同医学图像传递的信息可比,帮助我们从多个层面分析和诊断疾病。基于图像配准的学习和理解,以及对DCE-MRI(Dynamic Contrast-Enhanced Magnetic Resonance Image)和 fMRI(Functional Magnetic Resonance Image)图像本身特点的理解,本文致力于准确且合理地配准DCE-MRI和fMRI数据,为之后的医学图像分析提供更可靠的保证。归纳起来,本论文主要包括以下几个方面工作:1)造影剂的注入往往使得传统肝脏DCE-MRI配准方法产生不真实的形变。为了解决这个问题,我们提出了一种基于流形的配准方法。假设肝脏DCE-MR时间序列位于一个低维流形中,表示DCE-MR图像中帧与帧之间的本质相似度。基于这个流形,我们可以将两幅较不相似的图像之间的大形变场分解成流形上一系列相近图像的小形变场。这样,我们就可以沿着流形上的测地路径将每个图像配准到模板图像上。在配准过程中,我们使用鲁棒的主成分分解方法将造影剂导致的增强部分从原始图像中分离出来。同时也利用了求测地距离上的均值,自动选择出无偏差的模板图像。实验结果表明本文提出的方法在保证图像的拓扑结构的基础上,可以有效减少由于呼吸等因素造成的肝脏运动。2)使用静息态功能磁共振(resting-state fMRI)理解大脑功能需要将大脑功能区域准确地配准。通常都使用与fMRI对应的高分辨率结构图像的配准结果辅助fMRI配准。但是,这种配准方法并不能很好的对齐功能区域。因为解剖结构和功能区域不一定完全对齐。一些文献提出可以直接利用灰质或者大脑皮层上的功能信息对fMRI数据配准。由于灰质或者大脑皮层是一个薄且曲率高的结构,这样的功能信息很难得到全脑的功能区域的准确配准。越来越多的研究表明,大脑功能信息不仅仅存在于灰质中同时也存在于白质。基于这个事实,本文提出一种描述功能连接性各向异性的特征描述方法,不仅仅提取灰质的功能信息同时提取白质的功能信息,最终完成全脑的fMRI配准。本文新的配准方法主要分成两个部分:首先,利用组织特异的功能连接张量(tissue-specific patch-based functional connectivity tensors,ts-PFCTs),我们分别在白质和灰质上提取功能的灰度信息和方向信息。其次,使用多通道大形变微分同胚映射(multi-channel Large Deformation Diffeomorphic Metric Mapping,LDDMM)配准算法融合多组织特征得到fMRI准确的功能配准结果。实验结果表明与目前的方法相比,本文方法可以得到更好的功能一致性。
[Abstract]:The clinical application of medical images, such as disease monitoring and diagnosis, evaluation of healing results and so on, makes the research and analysis of medical images rapidly become a new field. Medical images can provide more and richer medical information, which requires timely and accurate interpretation and analysis. Therefore, medical image registration becomes very important and necessary. This method of processing allows different medical images to transmit information comparable, and helps us analyze and diagnose diseases on multiple levels. Learning and understanding based on image registration, as well as understanding the characteristics of DCE-MRI(Dynamic Contrast-Enhanced Magnetic Resonance image and fMRI(Functional Magnetic Resonance image, This paper is devoted to the accurate and reasonable registration of DCE-MRI and fMRI data, which provides a more reliable guarantee for medical image analysis. This paper mainly includes the following aspects: (1) injection of contrast media often results in unreal deformation of traditional liver DCE-MRI registration method. In order to solve this problem, In this paper, we propose a registration method based on manifold, assuming that the liver DCE-MR time series is located in a low dimensional manifold, which represents the essential similarity between frames in the DCE-MR image. We can decompose the large deformation field between two dissimilar images into small deformation fields of a series of similar images on the manifold. We can match each image to the template image along the geodesic path on the manifold. We use robust principal component decomposition to separate the contrast enhancement from the original image. We also use the mean value of the geodesic distance. The experimental results show that the proposed method is based on the topological structure of the image. Can effectively reduce liver movement caused by breathing and other factors. 2) use resting functional magnetic resonance imaging (fMRI) to understand brain function requires accurate registration of brain functional regions. High resolution structures corresponding to fMRI are usually used. Image registration results assist fMRI registration. However, This registration method can not align the functional regions very well, because the anatomical structure and functional areas are not always aligned completely. Some literatures suggest that the fMRI data can be registered directly by using the functional information of gray matter or cerebral cortex. Because gray matter or cortex is a thin, high-curvature structure, It is difficult to get accurate registration of functional regions in the whole brain. More and more studies show that the functional information of the brain exists not only in gray matter but also in white matter. In this paper, a feature description method for describing functional connectivity anisotropy is proposed, which not only extracts the functional information of gray matter, but also extracts the functional information of white matter. Finally, the fMRI registration of the whole brain is completed. The new registration method is divided into two parts: first, Using the tissue-specific patch-based functional connectivity connections of Zhang Liang and PFCTs, we extract gray and direction information from white matter and gray matter, respectively. The multi-channel Large Deformation Diffeomorphic Metric mapping algorithm for multi-channel Large Deformation Diffeomorphic Metric mapping LDDMMs is used to obtain the accurate functional registration results of fMRI. The experimental results show that the proposed method can achieve better functional consistency than the current methods.
【学位授予单位】：南方医科大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：R735.7;R445.2;TP391.41

【相似文献】