Graph-Based Medical Image Segmentation in 3D and 4D

描述(由申请人提供)：在体积数据集中有效地检测表示对象边界的全局最优表面是重要的，并且在许多医学图像分析应用中仍然具有挑战性。该方案解决了在3-D和4-D中检测最佳单个和多个相互作用的曲面的具体问题，包括圆柱形、闭合曲面形状和“复杂”形状。将开发允许在最佳表面检测框架中结合基于形状的先验知识的新方法。通过将三维图搜索问题转化为计算加权有向图中最优闭集的问题，实现了计算的可行性。将全局最优性与优化过程中使用的特定于问题的目标函数相结合，将有助于将该方法应用于各种医学图像分割问题。我们假设，基于最优图搜索的三维和四维表面检测的图像分割将在来自各种医学成像源的体图像数据中提供准确和稳健的分割性能，提供理论上的效率和实用上的适用性。1)提出并验证了一种适用于三维和四维(包括柱面和闭合曲面)生物医学图像分割的单个和多个交互曲面的最优检测方法。2)开发并验证了一种保持复杂拓扑结构的三维和四维最优表面检测方法。3)提出并验证了一种将形状先验信息融入到分割过程中的三维和四维最优曲面检测方法。开发的方法将与目前使用的最先进的方法进行比较测试。这些方法的性能将在足够大小的数据样本中进行统计评估。与公共卫生相关：体积图像扫描仪(例如，计算机断层扫描、磁共振、超声波)在医学上越来越普遍，但空间数据的分析通常是在逐个切片的基础上进行可视执行的。因此，大量的体积信息不能被医生充分利用。这里提出的图像分析方法允许以定量的方式客观地评估图像数据，有望对基于图像的临床护理产生重大影响。

DESCRIPTION (provided by applicant): Efficient detection of globally optimal surfaces representing object boundaries in volumetric datasets is important and remains challenging in many medical image analysis applications. This proposal deals with a specific problem of detecting optimal single and multiple interacting surfaces in 3-D and 4-D, including cylindrical shapes, closed-surface shapes, and "complex" shapes. Novel methods allowing incorporation of shape-based a priori knowledge in the optimal surface detection framework will be developed. The computational feasibility is accomplished by transforming the 3-D graph-searching problem to a problem of computing an optimal closed set in a weighted directed graph. Combining the global optimality with problem-specific objective functions used in the optimization process will facilitate application of the methods to a wide variety of medical image segmentation problems. We hypothesize that image segmentation based on 3-D and 4-D surface detection utilizing optimal graph searching will provide accurate and robust segmentation performance in volumetric image data from a variety of medical imaging sources, offering theoretical efficiency and practical applicability. We propose to: 1) Develop and validate a method for optimal detection of single and multiple interacting surfaces applicable to biomedical image segmentation in 3-D and 4-D (including cylindrical and closed surfaces). 2) Develop and validate a 3-D and 4-D optimal surface detection method that preserve complex topologies. 3) Develop and validate a 3-D and 4-D optimal surface detection method that incorporates shape priors into the segmentation process. The developed methods will be tested in comparison with state-of-the-art methods utilized today. The methods' performance will be statistically assessed in data samples of sufficient sizes. Public Health relevance: Volumetric image scanners (e.g., computed tomography, magnetic resonance, ultrasound) are increasingly available in medicine, yet the analysis of spatial data is typically performed visually on a slice-by-slice basis. The large amount of volumetric information therefore cannot be fully utilized by the physicians. Image analysis methods such as proposed here allow evaluating the image data objectively in a quantitative manner, promising to substantially impact image-based clinical care.