基于GPU的虚拟内窥技术研究

发布时间：2018-01-17 14:12

本文关键词：基于GPU的虚拟内窥技术研究　出处：《哈尔滨工业大学》2015年硕士论文　论文类型：学位论文

【摘要】：虚拟内窥技术为医疗诊断开拓了新方法,它是无创的、非侵入式的,可以检测极其微小的或传统内窥镜所无法探入的器官,如脑、心脏等。虽然目前虚拟内窥技术发展迅速,但仍存在中心路径提取速度慢、交互效率低等问题。因此,本文的主要研究目的是将GPU的并行计算性能融入到虚拟内窥技术的实现中,提高中心路径的提取效率,开发出具备高交互性能的虚拟内窥系统,为临床医学的诊疗提供帮助。在中心路径提取方面,本文采用势能场方法。为了降低该方法的时间复杂度,提出了基于GPU的势能场中心路径提取并行算法,并充分利用CUDA架构特有的常量存储器和共享存储器对普通并行算法进行改进。通过对多组3D模型进行测试,结果表明,随着数据规模的增大,加速效果逐渐提升。处理256×256×487的体数据时,可获得18倍的加速比,解决了中心路径提取时间长、效率低的问题。在三维可视化方面,本文实现了基于移动立方体算法的面绘制和基于光线投射算法的体绘制。由于体绘制运算量大,仅用CPU进行计算常常无法满足实时绘制的要求。因此,本文还提出了基于GPU的并行体绘制方法,利用GPU的并行计算性能来提高绘制帧率。测试结果表明,对于多组不同大小的3D模型,该并行方法的绘制帧率均可达到40以上,满足实时绘制的要求。另外,本文还提供了实时交互功能,用户可自行设置传递函数以达到理想的绘制效果,并可对可视化结果进行放缩、旋转和拖动等操作。在虚拟漫游方面,本文将平行投影改为透视投影以实现近大远小的视觉感官效果。提供引导式漫游方式,用户可操控虚拟相机沿着中心路径进行漫游,或固定在某一位置对局部进行细微观察,通过点击鼠标或按钮完成虚拟相机镜头的放缩或旋转等操作。最后,本文融合了以上三项技术,设计并实现了基于GPU的虚拟内窥系统,该系统具备较好的交互性能,可改善用户体验,为临床医学的诊疗提供有效的辅助平台。
[Abstract]:Virtual endoscopy opens up a new method for medical diagnosis. It is noninvasive and can detect extremely small or undetectable organs such as the brain. Although virtual endoscope technology is developing rapidly at present, there are still some problems such as slow center path extraction and low interaction efficiency. The main purpose of this paper is to integrate the parallel computing performance of GPU into the implementation of virtual endoscope, improve the efficiency of central path extraction, and develop a virtual endoscope system with high interaction performance. In order to reduce the time complexity of this method, a parallel algorithm based on GPU for center path extraction of potential energy field is proposed. And make full use of the CUDA architecture unique constant memory and shared memory to improve the common parallel algorithm. Through the multi-group 3D model test, the results show that, with the increase of data size. When processing 256 脳 256 脳 487 volume data, the acceleration ratio can be obtained by 18 times, which solves the problem of long time and low efficiency of center path extraction. In this paper, surface rendering based on moving cube algorithm and volume rendering based on ray-casting algorithm are realized. Because of the large amount of volume rendering computation, only CPU can not meet the requirements of real-time rendering. This paper also proposes a parallel volume rendering method based on GPU, which uses the parallel computing performance of GPU to improve the frame rate of rendering. The test results show that, for many groups of 3D models of different sizes. The rendering frame rate of the parallel method can reach more than 40, which can meet the requirement of real-time rendering. In addition, the real-time interactive function is provided in this paper. The user can set the transfer function to achieve the ideal rendering effect. In virtual roaming, the parallel projection is changed to perspective projection to realize the visual sense effect of near, far and small. The user can control the virtual camera to roam along the central path, or fix a certain position to make a fine observation of the local area, and by clicking the mouse or button to complete the virtual camera lens zooming or rotation operations. Finally. This paper integrates the above three technologies and designs and implements a virtual endoscope system based on GPU. The system has good interaction performance and can improve the user experience and provide an effective auxiliary platform for the diagnosis and treatment of clinical medicine.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R443;TP391.41

【参考文献】