数字脑—计算解剖学方法及GPU技术应用的研究
发布时间：2018-08-19 07:45
【摘要】： 随着人脑成像技术的发展,脑成像在脑科学和神经科学,神经外科等研究中具有越来越重要的地位。为对从个体人脑数据到人群中脑数据进行形态和功能的分析、比较,迫切需要研究数学方法和有效的计算手段,计算解剖学这一学科应运而生。针对人脑数据的计算神经解剖学重点研究的内容包括人脑图谱的建模、形变模型、结构和功能的映射分析等几大问题的数学和计算方法。本文主要研究了计算神经解剖学中若干个重要问题,即人脑图像中的结构和解剖标记点的提取、基于弹性模型的人脑形变配准技术、人脑皮层图像的解剖分块方法,最后研究了可编程GPU(图形处理器)技术在数字脑可视化和图像快速处理中的应用。 在人脑MRI图像的分割方面。针对脑与非脑组织的分割问题,分别使用较为简单的边界分割方法和基于分水岭的方法实现脑与非脑组织的分割。对脑组织的皮质分割,实现了一个基于概率图谱的模糊聚类方法,并研究了图像中的组织灰度不均匀性对分割的影响。在人脑解剖标记点提取方面,阐述了基于等值面曲率模型的数学方法,并实验验证半自动方法提取标记点,为后续的图像配准提供对应的解剖标记点。 对人脑图像的弹性配准的有限元计算方法首先阐述了其离散化的方法,并对二维图像的网格剖分提出了一个与图像特征相关的网格划分算法,使得网格具有一定的图谱特征。同时利用解剖标记点作为预先的图像刚性初配准,并作为有限元计算的位移条件加快计算的收敛速度和精度。 人脑皮层体数据图像是由脑的沟回组成的有复杂形态的解剖结构,对其按功能和解剖特征分块在fMRI数据分析和皮层脑沟回的自动识别方面都具有重要意义。我们提出了基于测地距离的K-均值空间聚类算法,提出聚类中心点的快速估计方法。从而实现了人脑皮层数据的与近似解剖特征的皮层分块。 为提高数字脑体绘制的成像质量和加快图像处理,我们实现了一套基于可编程GPU的可视化和图像处理的基本应用框架。提出了补偿体绘制质量的几种方法。对大规模体绘制问题,我们提出了基于矢量量化压缩后的体数据进行实时解码和绘制,从而为大规模体数据的绘制带来了新的基于硬件的快速方法。对GPU作为一种廉价的可并行计算的处理器,进行了一些并行图像处理方法的实现研究,如Level Set方法,骨架提取算法等。结果表明,采用GPU计算可以得到很好的计算加速性能。
[Abstract]:With the development of brain imaging technology, brain imaging plays a more and more important role in brain science, neuroscience, neurosurgery and so on. In order to analyze the morphology and function of human brain data from individual human brain data to human brain data, it is urgent to study mathematical methods and effective calculation methods, and the subject of computational anatomy emerges as the times require. The main contents of the research on the computational neuroanatomy of human brain data include the modeling of human brain atlas, deformation model, mapping analysis of structure and function, and so on. In this paper, some important problems in computational neuroanatomy, such as the extraction of structures and anatomical markers in human brain images, the technique of human brain deformation registration based on elastic model, and the anatomical segmentation of human cortical images are studied in this paper. Finally, the application of programmable GPU technology in digital brain visualization and image processing is studied. In the human brain MRI image segmentation. To solve the problem of brain and non-brain tissue segmentation, a simple boundary segmentation method and a watershed based method are used to segment brain and non-brain tissue, respectively. A fuzzy clustering method based on probabilistic map is implemented for cortical segmentation of brain tissue, and the effect of tissue grayscale heterogeneity on segmentation is studied. In the aspect of human brain anatomical mark point extraction, the mathematical method based on isosurface curvature model is expounded, and the semi-automatic method is verified by experiment, which provides the corresponding anatomical mark points for the subsequent image registration. The finite element method for the elastic registration of human brain images is introduced in this paper. Firstly, the discretization method is introduced, and a mesh generation algorithm related to the image features is proposed for the mesh generation of two-dimensional images, which makes the meshes have a certain graph feature. At the same time, the anatomic mark point is used as the initial registration of image rigidity, and the displacement condition calculated by finite element method is used to accelerate the convergence speed and accuracy of the calculation. The cortical body data image of human brain is a complex anatomical structure composed of the sulcus gyrus of the brain, which is of great significance in the analysis of fMRI data and the automatic recognition of the cortical gyrus according to its function and anatomical characteristics. We propose a K-means space clustering algorithm based on geodesic distance and a fast estimation method for clustering center points. Thus, the cortical block of human brain cortical data and similar anatomical features is realized. In order to improve the imaging quality of digital brain volume rendering and speed up image processing, we have implemented a set of basic application framework of visualization and image processing based on programmable GPU. Several methods of compensating volume rendering quality are presented. To solve the large-scale volume rendering problem, we propose a real-time decoding and rendering of volume data based on vector quantization compression, which brings a new fast method based on hardware for large-scale volume data rendering. As a cheap parallel computing processor, GPU is studied in the realization of some parallel image processing methods, such as Level Set method, skeleton extraction algorithm and so on. The results show that the acceleration performance can be obtained by GPU calculation.
【学位授予单位】：东南大学
【学位级别】：博士
【学位授予年份】：2005
【分类号】：R319;R322
本文编号：2191112
[Abstract]:With the development of brain imaging technology, brain imaging plays a more and more important role in brain science, neuroscience, neurosurgery and so on. In order to analyze the morphology and function of human brain data from individual human brain data to human brain data, it is urgent to study mathematical methods and effective calculation methods, and the subject of computational anatomy emerges as the times require. The main contents of the research on the computational neuroanatomy of human brain data include the modeling of human brain atlas, deformation model, mapping analysis of structure and function, and so on. In this paper, some important problems in computational neuroanatomy, such as the extraction of structures and anatomical markers in human brain images, the technique of human brain deformation registration based on elastic model, and the anatomical segmentation of human cortical images are studied in this paper. Finally, the application of programmable GPU technology in digital brain visualization and image processing is studied. In the human brain MRI image segmentation. To solve the problem of brain and non-brain tissue segmentation, a simple boundary segmentation method and a watershed based method are used to segment brain and non-brain tissue, respectively. A fuzzy clustering method based on probabilistic map is implemented for cortical segmentation of brain tissue, and the effect of tissue grayscale heterogeneity on segmentation is studied. In the aspect of human brain anatomical mark point extraction, the mathematical method based on isosurface curvature model is expounded, and the semi-automatic method is verified by experiment, which provides the corresponding anatomical mark points for the subsequent image registration. The finite element method for the elastic registration of human brain images is introduced in this paper. Firstly, the discretization method is introduced, and a mesh generation algorithm related to the image features is proposed for the mesh generation of two-dimensional images, which makes the meshes have a certain graph feature. At the same time, the anatomic mark point is used as the initial registration of image rigidity, and the displacement condition calculated by finite element method is used to accelerate the convergence speed and accuracy of the calculation. The cortical body data image of human brain is a complex anatomical structure composed of the sulcus gyrus of the brain, which is of great significance in the analysis of fMRI data and the automatic recognition of the cortical gyrus according to its function and anatomical characteristics. We propose a K-means space clustering algorithm based on geodesic distance and a fast estimation method for clustering center points. Thus, the cortical block of human brain cortical data and similar anatomical features is realized. In order to improve the imaging quality of digital brain volume rendering and speed up image processing, we have implemented a set of basic application framework of visualization and image processing based on programmable GPU. Several methods of compensating volume rendering quality are presented. To solve the large-scale volume rendering problem, we propose a real-time decoding and rendering of volume data based on vector quantization compression, which brings a new fast method based on hardware for large-scale volume data rendering. As a cheap parallel computing processor, GPU is studied in the realization of some parallel image processing methods, such as Level Set method, skeleton extraction algorithm and so on. The results show that the acceleration performance can be obtained by GPU calculation.
【学位授予单位】：东南大学
【学位级别】：博士
【学位授予年份】：2005
【分类号】：R319;R322
【相似文献】
相关期刊论文 前10条
1 吕彬;何晖光;吕科;赵明昌;张志强;卢光明;;脑皮层厚度分析方法及其应用[J];中国医学影像技术;2008年07期
2 ;[J];;年期
3 ;[J];;年期
4 ;[J];;年期
5 ;[J];;年期
6 ;[J];;年期
7 ;[J];;年期
8 ;[J];;年期
9 ;[J];;年期
10 ;[J];;年期
相关博士学位论文 前1条
1 吴仲乐;数字脑—计算解剖学方法及GPU技术应用的研究[D];东南大学;2005年
，本文编号：2191112
本文链接：https://www.wllwen.com/yixuelunwen/binglixuelunwen/2191112.html
最近更新
教材专著
