Robust Cerebrum and Cerebellum Segmentation for Neuroimage Analysis

描述（由申请人提供）：从结构磁共振图像中提取大脑和小脑是神经图像分析的重要初始步骤。不准确的脑提取（也称为颅骨剥离）会对随后的分析产生非常负面的影响，对于一些敏感的研究，人工辅助提取仍然是唯一可行的选择。尽管有许多算法和一些软件包可用于研究目的，但这些方法的性能仍然存在广泛的可变性，并且没有一个适合于涉及整个人类大脑和大量研究的综合分析。该R21将为神经科学社区开发、编码、测试和分发SPECTRE（颅外组织去除的简单范例）软件。根据项目公告PAR-08-183，这项与NA-MIC国家生物医学计算中心的探索性合作将作为R21项目获得资助，在“NA-MIC工具包”软件环境下共同开发SPECTRE软件，并将其作为源代码和平台特定可执行文件免费提供。底层SPECTRE图像处理算法最近在约翰霍普金斯大学的图像分析和社区实验室（IACL）开发和验证，并已在一个主要会议上报告。SPECTRE在使用多个地图集、模糊分类、分水岭分割和形态学图像分割的组合使用以及强调意外去除皮质灰质的高惩罚方面具有创新性。研发工作将实现以下具体目标：1)现有代码和所有新代码将使用标准的NA-MIC软件方法进行移植和编写；2)完成小脑坐标系统的隔离和建立算法；3)新代码将针对不同采集的t1加权数据进行测试和优化；4)将SPECTRE与现有算法进行广泛的比较。结果将是一种算法，它可以取任意的t1加权MR脑体积，并返回包含小脑、大脑或两者的体积，并在大脑和小脑上自动建立坐标系统。SPECTRE软件工具将是第一个提供选择性分离大脑、小脑或两者的工具。这也将是第一个在小脑上建立坐标系统的自动化方法。应该指出的是，我们开发的优化标准是专门为非常敏感的大脑变化研究而设计的，这导致我们在隔离步骤中对错误的灰质损失进行了非常高的惩罚。由于这个原因，也因为软件工具将非常强大，易于使用，并将在广泛使用的3D切片机软件的丰富指定环境中运行，我们预计SPECTRE软件工具将在神经成像社区中变得非常受欢迎。与公共卫生相关：脑科学的许多进展都是利用大脑的磁共振图像发现的。这些数据的自动处理，通常在任何给定的研究中涉及大量的受试者，是获得科学或医学知识的必要组成部分，而大脑的自动识别通常是该过程中关键的第一步。这个研究项目将提供软件，免费向公众开放，自动识别人类大脑的大脑和小脑，以便进行进一步的分析，无论是传统的还是潜在的新分析。

DESCRIPTION (provided by applicant): Extraction of the cerebrum and cerebellum from structural magnetic resonance images is an important initial step in neuroimage analysis. Inaccurate brain extraction (also referred to as skull stripping) can have a very negative effect on subsequent analyses, and for some sensitive studies, manually-assisted extraction remains the only viable option. Despite many reported algorithms and several software packages available for research purposes, there is still widespread variability in the performance of these methods, and none are adequate for comprehensive analyses involving the whole human brain and large numbers of studies. This R21 will develop, code, test, and distribute the SPECTRE (Simple Paradigms for Extra Cranial Tissue REmoval) software for the neuroscience community. Funded as an R21 under the Program Announcement PAR-08-183, this Exploratory Collaboration with the NA-MIC National Center for Biomedical Computing, will jointly develop SPECTRE software within the "NA-MIC Kit" software environment and will make it freely available as both source code and platform specific executables. The underlying SPECTRE image processing algorithm was recently developed and validated in the Image Analysis and Community Laboratory (IACL) at Johns Hopkins University and has been reported in a leading conference. SPECTRE is innovative in its use of multiple atlases, its combined use of fuzzy classification, watershed segmentation, and morphological image segmentation, and its emphasis on a high penalty for accidentally removing cortical gray matter. Research and develop efforts will accomplish the following specific aims: 1) The existing code and all new code will be ported and written using the standard NA-MIC software methodology; 2) The algorithms for isolating and establishing a coordinate system on the cerebellum will be completed; 3) The new code will be tested and optimized for differently acquired T1-weighted data; 4) An extensive comparison between SPECTRE and existing algorithms will be carried out. The result will be an algorithm that can take an arbitrary T1-weighted MR brain volume and return a volume containing the cerebellum, the cerebrum, or both, and with coordinate systems automatically established on the cerebrum and cerebellum. The SPECTRE software tool will be the first to provide selective isolation of the cerebrum, the cerebellum, or both. It will also be the first automated method for establishing a coordinate system on the cerebellum. It should be noted that the optimization criteria we have developed is particularly designed for very sensitive studies of brain changes, which has led us to incorporate a very high penalty on erroneous gray matter loss during the isolation step. For this reason and also because the software tool will be very robust, easy to use, and will function within the richly appointed environment of the widely used 3D Slicer software, we expect that the SPECTRE software tool will grow to be very popular within the neuroimaging community. PUBLIC HEALTH RELEVANCE: Many advances in brain science are discovered using magnetic resonance images of the brain. Automatic processing of these data, often involving very large numbers of subjects in any given study, is a necessary component in gaining scientific or medical knowledge, and automatic identification of the brain is typically a key first step in the process. This research project will provide software, freely available to the public that will automatically identify the cerebrum and cerebellum of the human brain so that further analysis, both conventional and potentially novel, can then be carried out.