SEMI-AUTOMATED METHODS OF SEGMENTATION OF BRAIN IMAGE

脑图像分割的半自动方法

• 批准号: 3745210
• 负责人: U RUTTIMANN
• 金额: --
• 依托单位: NATIONAL INSTITUTE ON ALCOHOL ABUSE AND ALCOHOLISM
• 依托单位国家: 美国
• 资助国家: 美国
• 起止时间: 至
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/3745210
• 关键词: brain imaging /visualization /scanning; brain mapping; computed axial tomography; computer assisted diagnosis; computer system design /evaluation; diagnosis design /evaluation; diagnosis quality /standard; human subject; image processing; magnetic resonance imaging; positron emission tomography

In order to achieve three-dimensional co-registration of images acquired with different technologies, corresponding landmarks must be identified in the respective images. The scale and orientation differences associated with image representations from different scanners present a major obstacle in this task. Visual identification of corresponding regions introduces subjectivity and inconsistency, and may become too laborious for large study series. Therefore, the aim of this research is the development of semi- or fully-automated computer methods for the identification and delineation ("segmentation") of important areas and landmarks (e.g. skull, gray and white brain matter,CSF, pathologic tissues) in brain images derived from different modalities. A fully-automated ("unsupervised") procedure to segment CT images into regions of bone, brain parenchyma, and CSF has been implemented. Calibration of this procedure with a CT image of an anthropomorphic phantom demonstrated unbiased segmentation. Its application for the segmentation of CSF spaces in alcoholic and normal subjects yielded results consistent with a subjective segmentation. For the purpose of enabling the registration of images that represent local glucose utilization (PET) with structural images (MRI or CT), algorithms for the detection of anatomic landmarks in PET slices were developed. Reliable automated detection of the outlines of the skull and the midlines in each slice was achieved, demonstrating suitability of these landmarks for registration.

为了实现获得图像的三维共同注册 使用不同的技术，必须在 各自的图像。 相关的规模和方向差异 带有来自不同扫描仪的图像表示形式 这项任务的障碍。 相应区域的视觉识别 引入主观性和不一致，可能变得太艰辛 大型学习系列。 因此，这项研究的目的是 开发半自动化的计算机方法 重要领域的识别和描述（“分割”） 地标（例如头骨，灰色和白色脑物质，CSF，病理学 组织）在源自不同方式的大脑图像中。 一个 完全自动化（“无监督”的）程序，以将CT图像分割为 已经实施了骨骼，脑实质和CSF的区域。 使用拟人化的CT图像对此过程进行校准 幻影表现出无偏分段。 它的应用 酒精和正常受试者中CSF​​空间的分割 结果与主观分割一致。 目的 启用代表局部葡萄糖的图像的注册 具有结构图像（MRI或CT）的利用率（PET），算法 开发了宠物切片中解剖地标的检测。 可靠的 自动检测颅骨和中线的轮廓 实现了切片，证明了这些地标的适用性 登记。