CASIA OpenIR  > 毕业生  > 博士学位论文
面向复杂医学图像的GPU体绘制引擎研究
Alternative TitleGPU Volume Rendering Engine for Complex Medical Image Data
杨飞
Subtype工学博士
Thesis Advisor田捷
2014-05-25
Degree Grantor中国科学院大学
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医学影像 体绘制 可视化 Gpu 光学模型 Medical Imaging Volume Rendering Visualization Gpu Optical Model
Abstract近代医学影像技术的提出使得医生可以在无创伤的情况下观察病人体内的变化。作为出于临床诊断的需求提出的技术,从诞生开始,医学影像技术的核心任务之一就是对医学图像的显示。随着计算机技术的发展,数字化进入医学影像技术的标准流程。把医学图像转化为便于处理的数据形式,给医学图像的显示带来了更多的可能性。为此,计算机图形学领域的建模技术和绘制技术被引入医学影像领域,形成了独特的交叉方向:医学影像可视化。由于目前医学影像数据自身的特点,医学影像的可视化主要是对三维体数据的可视化。这类可视化技术除了在医学影像领域之外,还在无损检测,科学仿真等领域有重要的应用。然而与其他应用相比,医学影像可视化的突出特点是数据的复杂性。医学影像成像对象――生物体,具有固有的复杂性。在结构方面,不同的组织交错、缠绕在一起;在属性方面,生物组织具有多面性,需要从不同角度加以认识。为了全面、精确地检测生物体的各种参数,成像设备目前正向着多维度、多模态、高精度的方向发展,由此产生的数据具有相应的复杂性。同时,计算机技术也按照其内在规律向前发展,由串行计算占主流发展为现在并行计算无处不在的局面,尤其是GPU技术的普及,使得过去一些传统的技术方法受到挑战,需要重新设计。 本文工作针对医学影像数据特有的复杂性和最新的高性能GPU的技术特点,对医学影像可视化的核心技术――体绘制引擎展开深入研究,一方面从计算框架的角度研究适合于GPU的模块化的算法设计方法,另一方面针对医学影像数据在结构和属性两类复杂性提出多项技术改进。综上,本文工作共包括三个部分: 1. 设计新的体绘制引擎计算框架。新的框架一方面要能充分发挥出当前商品市场上主流硬件的计算性能,尽可能提高体绘制引擎的实时流畅性,另一方面要能满足复杂医学图像可视化的多样化需求。为此,本文首先提出一个基于静态连接的三层可复用框架,继而利用最新的OpenCL动态编译技术提出一个动态可复用框架,并将二者集成在统一的计算平台之上,可根据需要来选择。 2. 针对医学影像数据在结构方面的复杂性,提出光线特征分析和等值面分割两种结构分析技术。光线特征分析可以用来在不进行分割的情况下快速地对体数据中的组织进行分类。等值面分割可以在三维的等值面场景中交互地对组织器官进行选择分割,并立即显示。 3. 针对医学影像数据在属性方面的复杂性提出一个底层光学模型和一个多体数据模型绘制系统。底层光学模型的改进主要是为了更好地刻画生物体复杂的内部成分,将三维物体看成是由颗粒物质和表面结构混合在一起的整体,对于致密组织,增加表面反射的成分,对于软组织,则增加颗粒散射的成分。多体数据模型绘制基于动态可复用框架,可以动态地将多个体数据和裁剪体加入同一场景当中进行GPU绘制,实现对具有不同属性的多个数据的同时绘制。 全文共分五章。第一章对医学影像可视化的引擎和预处理技术的研究背景、意义及国际上的研究现状进行介绍;第二章介绍面向医学影像需求的体绘制引擎的计算框架设计和实现;第三章介绍针对医学影像数据结构复杂性的技术研究;第...
Other AbstractThe introduction of medical imaging enables the noninvasive observation of the internal abnormal of patients. With the development of computer science, digitalization of medical images becomes a standard procedure of modern medical imaging. Since the technology is driven by the need of clinical diagnosis, the display of images is a core task of medical imaging from the beginning. The conversion of images into digital data brings more possibilities. Therefore, the modeling and rendering technologies are introduced into the medical imaging domain, which forms a special interdisciplinary: medical image visualization. Due to the natural of current medical image data, medical image visualization is mainly designed for 3D volume data. This class of visualization technologies is also used in non-medical applications such as nondestructive testing and scientific simulations. Comparing to these applications, complexity is a notable feature of medical imaging. The subjects of medical imaging are organisms, where intrinsic complexities exist. There are several kinds of complexities. In the structural aspect, different organs wind together. In the attribute aspect, organs have different natures and need to be investigated from different aspects. To detect the parameters of organisms in a thorough and accurate way, multi-dimension, multi-modality, and high-precision features are more and more commonly seen in recent medical imaging devices. Inherently, the resulted image data get more and more complicated. At the meantime, computer science is under significant development. It has not been a very long time since sequential computing dominated the consumer market. Now, parallel computing is everywhere. Especially since GPUs became prevalent, some conventional designing paradigms are being challenged. This work focuses on the unique complexity of medical image and addresses the new features of current high-performance GPU platforms. In this work, GPU based volume rendering techniques will be investigated from several aspects. On one hand, new frameworks are developed to facilitate modularized GPU algorithm development. On the other hand, to address the complexity of medical image data, several technical impartments are proposed. Thus, the work contains 3 parts: 1. Develop new frameworks for volume rendering engine. The new frameworks should utilize the current commodity hardware to their full-strength so that the volume rendering engines can achieve the best performa...
shelfnumXWLW2024
Other Identifier201118014628064
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/6595
Collection毕业生_博士学位论文
Recommended Citation
GB/T 7714
杨飞. 面向复杂医学图像的GPU体绘制引擎研究[D]. 中国科学院自动化研究所. 中国科学院大学,2014.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20111801462806(4183KB) 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
[杨飞]'s Articles
Baidu academic
Similar articles in Baidu academic
[杨飞]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[杨飞]'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.