《基于体绘制方法的医学图像反分割方法研究与应用稿》由会员分享，可在线阅读，更多相关《基于体绘制方法的医学图像反分割方法研究与应用稿（64页珍藏版）》请在金锄头文库上搜索。
1、基于体绘制的医学图像反分割方法研究与应用专业名称：计算机应用技术 申请者名称：陈彦达 导师姓名：鲍苏苏摘摘 要要医学图像分割是得到人体组织、器官以及病变体的三维图像、仿真手术等后续处理的基础，它在医学影像处理与分析中具有特殊重要的意义，它是医学图像处理的关键一步，是个跨越医学和计算机科学综合性研究课题。因此，对医学图像分割的研究，具有重要的学术意义和应用价值。本文首先介绍了课题研究的背景与意义，概述了国内外医学图像分割的发展现状和趋势，以及本论文的主要研究内容和框架，对当前一些常用的医学图像分割算法和三维重建算法进行了分析和实验。其次，由于传统的图像分割过程都是走先分割再重建的流程。因此，存在
2、着下面的一些问题：分割结果的三维图像需要到重建后才能得到，这样 就无法及早发现分割过程中所出现的问题；无法观察分割过程或与之进行交互；分割算法所需要的种子点无法准确 获得 。为了克服上述困难，本文提出反分割方法，实现了获取三维种子点；分割算法与三维体绘制算法的结合与同步显示；分割与重建迭代执行与交互。通过实验明确了该方法的可行性与可用性。为了 应用 于反分割方法 ，提高区域生长的分割精度，减少种子点选取对分割结果的影响和用户交互量，本文提出了一种改进的快速区域生长分割算法。 该算法 通过置信区间和区域竞争计算目标区域最优阈值范围 。在方法上区域生长方法考虑的是图像的局部信息，而置信区间和区域竞
3、争方法考虑的是图像的全局信息。本文的分割算法融合了两者的优点。通过在一张图片上选择目标对象和背景对象的多个种子点，实现了复杂背景下的序列图像分割。使用一组腹部CT 原始图片进行的实验结果表明，算法在只需很少交互的情况下，有效地提高了分割精度。由于 体数据 三维重建 规模较大 ，大部分体绘制算法都难以实现交互的动态绘制。为解决医学图像三维可视化中大规模体数据显示速度与成像质量问题，引入一种基于蒙特卡罗方法的体绘制算法。该算法根据给出的概 率密度函数从随机样本点或其子集中选取一个点阵来进行绘制，适用于 大规模体数据集的有效可视化。另外，使用渐进细化的方法在图像质量和交互性上进行折衷，以适应实时要求
4、。实验结果表明，在最终图像分辨率固定下，该算法的时间复杂度和空间复杂度都比之前的算法要好，成像质量也满足实时绘制要求。最后，对整个论文的工作进行了总结与分析，并指出了存在的一些研究难点，技术难点和进一步研究的方向。 关键词：区域生长；区域竞争；置信区间；体绘制；蒙特卡罗积分；三维种子点A INVERSE SEGMENTATION METHOD OF MEDICAL IMAGES BASE ON VOLUME RENDERINGMajor: Computer Application Technology Name: Yan-da Chen Supervisor: Su-su BAOABSTRAC
5、TIn medical field, segmentation for medical image can obtain 3D image of organ which are widely used in diagnosis, surgery planning and simulating. Image segmentation is a key step in image processing. It is also a synthetic research task which is related to medical and computer science. With the ra
6、pid development of medicine, image segmentation is taking an important role in medical application.First, the background and the significance of this topic research are introduced. After making the analysis of the development present situation and the tendency of domestic and foreign medicine image
7、segmentation, main research content and frame of this article have been outlined. And the hot algorithms of image segmentation and reprocessing in current time have been introduced.Next, the traditional process of segmentation is that doing reprocessing after image segmentation. So, there are some p
8、roblems. Such as, doctor can not see the 3D result of segmentation before reprocessing, he can also not to interact with the process of segmentation, and the seeds of segmentation algorithm cant be got correctly. To solve these problems, the inverse segmentation is presented. The innovations of this
9、 process are following: Getting the 3D seeds in volume rendering, the algorithm of segmentation and reprocessing have been combinated and displayed synchronal. The feasibility and usability of this process have been proved by some experimentation.In order to apply by inverse segmentation, improve th
10、e accuracy of segmentation of region growing and reduce the number of user interaction, a segmentation algorithm of region growing based on confidence interval and region competition is presented. Region growing method focuses on local variations of an image while confidence interval and region comp
11、etition can extract a global property of an image. This approach combines both advantages. And the segmentation from image-sequences of complex background can be achieved by selecting some seeds from object and background in an image of image-sequences. The experimental results with a serial of abdo
12、minal CT images show that the proposed algorithm can improve the accuracy of segmentation effectively with only very little interactionFrom the aspect of 3D reprocessing, due to the large data sets, most of current algorithms are difficult to achieve a dynamic interactive rendering. In order to reso
13、lve the rendering speed and imaging quality problem in 3D Visualization of medical image, a novel volume rendering algorithm based on Monte Carlo method is presented. According to the given probability density function, a point cloud of random samples or its subset is selected and then rendered. The
14、 algorithm is mainly presented to efficiently visualize the large volume data sets. And the trade-off between image quality and interactivity can be obtained by using progressive refinement. The experiment results show that given a fixed image resolution the time complexity and the memory complexity
15、 of the algorithm are better than previous methods. And image quality meets the demand of real-time rendering.Finally, after summary and analysis to the whole paper work, some difficulties and further studies direction have been pointed out.KEY WORDS：region growing; region competition; confidence interval; Volume rendering; Monte Carlo; integration; 3D Seed point目目 录录摘 要.IABSTRACT.III目 录.VI第一章 绪 论.
