DVMT OF SEGMENTATION METHODS TO EXTRACT ANATOMIC FEATURES FROM BRAIN IMAGING

从脑成像中提取解剖特征的分割方法的 DVMT

• 批准号: 6123553
• 负责人: William Eric Grimson
• 金额: $ 2.8万
• 依托单位: BRIGHAM AND WOMEN'S HOSPITAL
• 依托单位国家: 美国
• 财政年份: 1998
• 资助国家: 美国
• 起止时间: 1998-09-30 至 1999-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/6123553
• 关键词: bioengineering /biomedical engineering; biomedical resource; nervous system; technology /technique

We plan to develop a novel software environment that will allow us to prepare magnetic resonance images (MRI) and computer assisted tomograms (CT) for quantitative image analysis and for intraoperative guidance. In order to accomplish our goals, we need to: consolidate already available image processing software infrastructure with the high performance computational hardware and networking facilities; improve and optimize our operator controlled segmentation environment (OCS); develop existing signal intensity based segmentation (SIS); develop existing template driven segmentation algorithms (TDS); develop an objective methodology for validation of the algorithms. OCS tools will include the development of a volume editor (VE). The second, SIS tools, will include the addition of Markov Random Fields (MRF) and Mean Fields (MF) to be added and integrated with our current expectation maximization (EM) segmenter. MRF will allow us to more fully exploit the image data by letting neighboring voxels influence the decision about the tissue class of a given voxel. This will be particularly helpful in brain regions where there is overlap between tissue classes. The inclusion of MF will enable larger neighborhoods to be taken into account in making these determinations. Thus both MRF and MF will serve as boosters to our current EM segmenter. And third, TDS tools will be developed in order to identify neuroanatomical regions of interest both quickly and more precisely. Here, we will build upon our existing brain atlas data set by adding more regions of interest. We will also use linear registration techniques as a first step in registering new brains to our atlas. The next step will involve elastic registration which is currently available in a single processor version, but will be migrated to our parallel processing machines in order to increase the number of cases that can be completed in the shortest period of time. In addition, because the brain atlas is based on a normal human brain, we will develop tools which will be used to identify, and to measure, the size and location of abnormal tissue such as brain tumors.

我们计划开发一个新颖的软件环境，这将使我们 准备磁共振图像（MRI）和计算机辅助 断层图（CT）用于定量图像分析和术中 指导。 为了实现我们的目标，我们需要：巩固 可用的图像处理软件基础架构 性能计算硬件和网络设施；提升 并优化我们的操作员受控分割环境（OCS）； 开发现有的基于信号强度的分割（SIS）；发展 现有模板驱动分段算法（TDS）;开发一个 验证算法的客观方法。 OCS工具将包括开发卷编辑器（VE）。 这 其次，SIS工具将包括添加马尔可夫随机字段 （MRF）和平均字段（MF）将与我们的当前添加并集成 期望最大化（EM）细分器。 MRF将允许我们更多 通过让相邻的体素影响到完全利用图像数据 关于给定体素的组织类别的决定。 这将是 在大脑区域中特别有用的大脑区域之间存在重叠 组织类别。 MF的包含将使更大的社区 要考虑这些决定。 因此，这两者 MRF和MF将作为我们当前EM细分器的助推器。 第三，将开发TDS工具以识别 迅速，更精确的神经解剖区域既迅速又更精确。 在这里，我们将通过添加我们现有的大脑图集数据集建立 更多的关注区域。 我们还将使用线性注册 技术是向我们的地图集注册新大脑的第一步。 下一步将涉及当前的弹性注册 单个处理器版本可用，但将迁移到我们的 并行处理机以增加案件数量 可以在最短的时间内完成。 此外， 因为大脑地图集是基于正常的人脑，所以我们将 开发将用于识别和测量大小的工具 和异常组织（例如脑肿瘤）的位置。