IMAGE PROCESSING AND GEOMETRICAL MODELING

本子项目是利用由NIH/NCRR资助的中心赠款提供的资源的众多研究子项目之一。子项目和研究者（PI）可能已经从另一个NIH来源获得了主要资金，因此可以在其他CRISP条目中表示。列出的机构是中心的，不一定是研究者的机构。本技术子项目涉及科学和医学数据的处理或分析问题。根据数据的类型和应用程序的目标，数据处理的技术水平各不相同。例如，信号处理领域，我们在这里指的是一维函数或波形的分析，是比较成熟的。信号处理仍然是重要的研究课题，但是有各种众所周知的、有效的、通用的算法来滤波和分类信号。具体的应用有很多，从语音识别到心脏监测。图像是多维信号，即在二维、三维或高维域上定义的函数。图像处理领域较年轻，而且已被证明更具挑战性。图像的重要方面不仅在其灰度（或光谱）值中编码，而且在其描述的形状中编码。例如，当考虑MRI数据时，皮层不仅由其强度定义，还由其形状和与其他解剖结构的空间关系定义。研究人员正在开发有效的图像分析技术，但这些技术还远远不够成熟，尚未在生物医学科学家群体中广泛采用。几何是指在空间中组织形成流形的点的集合。与信号和图像不同，几何对象（或流形）不一定是函数。这些流形所处的空间可以是二维的、三维的或n维的（其中n是3）。此外，这些点可以以不同的方式组织形成曲线、曲面、超曲面或由这些其他对象的组合组成的更复杂的对象。数字曲面的几何处理是一个相对年轻的领域，还有许多理论和实践问题有待解决。例如，表示数字曲面的问题本身就相当复杂，研究人员仍在研究各种可能性，包括点集、网格、多项式补丁和隐式曲面。几何处理既包括对几何对象的分析，也包括从科学数据中生成几何模型。该项目涉及生物医学应用的图像和几何处理。我们将把信号处理视为一种成熟的技术，我们将通过与其他工具包集成并依靠合作者的工作将其纳入我们的应用程序。我们在图像和几何处理方面的研究和开发目标将反映这些技术的相对成熟度，它们目前对生物医学研究人员的可用性，科学计算与成像研究所和我们的合作者的专业知识，以及驱动应用的特定需求。图像和几何处理的领域是广阔的，各种生物研究人员的数据处理需求是广泛的。该中心与该技术领域相关的资源相对较少，如果我们将它们与整个领域进行比较，甚至与其他正在进行的项目和专注于图像分析（例如）的中心进行比较的话。考虑到这一点，我们已经为这个核心采用了一种策略，利用犹他大学和其他地方正在进行的图像和几何处理研究，并扩展这项工作，以解决阻止我们的合作者充分利用最先进技术的具体障碍。鉴于此，我们的主要目标是解决可用性和可伸缩性问题。解决这些问题需要进行一些基础研究，但也需要与本提案中的其他技术核心紧密结合，并与在生物领域工作的其他团队进行重要合作。该项目的具体目标分为两组：研究目标和发展目标。研究目标是那些我们期望在算法层面上有一些基础工作或广泛工程的目标。这也意味着一些尚未经过相关应用测试的方法的开发，这意味着一些风险或一些潜在的算法重做。开发目标是指开发已知算法的新实现和将算法集成到新系统中。然而，它还包括在专门的计算架构上开发更快的实现，并在某些情况下包括对有很高成功可能性的并行算法的研究。图像处理的研究目标：(1)鲁棒滤波方法：开发新的，更通用的图像滤波方法，可以更容易地应用于更广泛的应用，减少自由参数的调整。(2)不完整和噪声层析成像数据集的分割：对重建伪影具有鲁棒性的电子显微镜层析成像数据集的自动和半自动分割方法。(3)用户交互分割：细化分割算法，使其与二维和三维可视化功能有效交互。几何处理的研究目标：(1)统计形状表征：形状变形的统计表征公式，适用于大型，铰接的解剖模型。(2)随机模式生成：数据驱动的中尺度模式生成，用于模拟微观（如细胞）结构的聚集效应。图像处理的发展目标：(1)并行实现：并行（分布式和共享内存）实现滤波和配准的迭代算法。(2)基于阿特拉斯的头部分割：将基于阿特拉斯的头部分割整合到模型生成和脑电源定位管道中。(3)主动形状模型（asm）：扩展自适应形状模型的ITK实现，包括：对三维模型的支持，与半自动分割方法（例如，分水岭和水平集模型）的集成，以及分层铰接模型。几何处理的发展目标：(1)网格生成：二维和三维网格生成，包括包含特定应用几何约束的四面体和六面体网格。(2)手动网格编辑：用户指导的网格几何形状和拓扑的操作。(3)基于点的配准：将基于点/曲线/曲面的配准算法集成到反问题工作流中。

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. This technical sub project deals with the problem of processing or analyzing scientific and medical data. The state of the art for data processing varies, depending on the type of data and goals of the application. For instance, the field of signal processing, which we use here to refer to the analysis of one-dimensional functions or waveforms, is somewhat mature. Important research topics remain in signal processing, but there are a variety of well-known, effective, general algorithms for filtering and classifying signals. Specific applications abound, from speech recognition to cardiac monitoring. Images are multidimensional signals, that is, functions defined on two-dimensional, three-dimensional, or higher- dimensional domains. The field of image processing is younger, and it has proven to be more challenging. The important aspects of images are encoded not only in their grey-scale (or spectral) values, but in the shapes that they describe. For instance, when considering MRI data, the cortex is defined not simply by its intensities but also by its shape and its spatial relationships to other anatomy. Researchers are developing effective technologies for image analysis, but the techniques are far from mature and have not yet been widely adopted within the community of biomedical scientists. Geometry refers to collections of points that are organized in space to form manifolds. Unlike signals and images, geometric ob jects (or manifolds) are not necessarily functions. The space in which these manifolds live could be two- dimensional, three-dimensional, or n-dimensional (where n > 3). Furthermore these points can be organized in different ways to form curves, surfaces, hypersurfaces, or more complex ob jects that consist of combinations of these other ob jects. Geometry processing for digital surfaces is a relatively young field, and a great many theoretical and practical questions remain. For instance, the problem of representing digital surfaces is itself quite complex, and researchers are still investigating a variety of possibilities including point sets, meshes, polynomial patches, and implicit surfaces. Geometry processing, includes both the analysis of geometric ob jects and the generation of geometric models from scientific data. This project addresses the processing of images and geometry for biomedical applications. We will consider signal processing as a somewhat mature technology, and we will include it in our applications by integrating with other toolkits and relying on the work of our collaborators. Our research and development aims in image and geometry processing will reflect the relative maturity of each of these technologies, their current availability to biomedical researchers, the expertise of the Scientific Computing and Imaging Institute and our collaborators, and the specific needs of driving applications. The fields of image and geometry processing are vast, and the data processing needs of various biological researchers are extensive. The Center¿s resource associated with this technical domain are relatively small¿if we compare them to either the field as a whole or even to other ongoing pro jects and centers that focus more exclusively on image analysis (for instance). With this in mind, we have adopted, for this core, a strategy of leveraging ongoing research in image and geometry processing, at Utah and elsewhere, and extending this work to address the specific roadblocks that prevent our collaborators from taking full advantage of state-of-the-art technologies. In light of this, we have focused the aims to address primarily issues of usability and scalability. Addressing these issues wil l entail some fundamental research, but it will also entail a tight integration with other technical cores in this proposal and significant collaborations with other teams working in biological areas. The specific aims of this project are divided into two groups: research goals and development goals. The research goals are those for which we expect there will be some fundamental work or extensive engineering at the algorithm level. It also implies some development of methods that have not been tested for the associated applications¿implying some risk or some potential reworking of algorithms. The development goals refer to the development of new implementations of known algorithms and the integration of algorithms into new systems. However, it also includes the development of faster implementations on specialized computing architectures and includes, in some cases, the investigation of parallel algorithms for which there is a high likelihood of success. Research Goals for Image Processing: (1) Robust Filtering Methods: The development of new, more general methods for image filtering that can be more easily applied across a wide range of applications with less tuning of free parameters. (2) Segmentation of Incomplete and Noisy Tomographic Datasets: Methods for automatic and semiautomatic segmentation of electron microscope tomography datasets¿robust to reconstruction artifacts. (3) User-Interactive Segmentation: The refinement of segmentation algorithms to interact effectively with two- dimensional and three-dimensional visualization capabilities. Research Goals for Geometry Processing: (1) Statistical Shape Characterization: Formulations for the statistical characterization of shape deformations with applicability to large, articulated anatomical models. (2) Stochastic Model Generation: The data-driven generation of mesoscale models for simulation of aggregate effects of microscopic (e.g., cellular) structures. Development Goals for Image Processing: (1) Parallel Implementations: Parallel (distributed and shared memory) implementations of iterative algorithms for filtering and registration. (2) Atlas-Based Head Segmentation: Integrate atlas-based head segmentation into model generation and EEG source localization pipeline. . (3) Active Shape Models (ASMs): Extend the ITK implementation of adaptive shape models to include: support for three- dimensional models, integration with semi-automated segmentation methods (e.g., watersheds and level-set models), and hierarchical articulated models. Development Goals for Geometry Processing: (1) Mesh Generation: Two-dimensional and three-dimensional mesh generation, including tetrahedral and hexahedral meshes that incorporate application-specific geometric constraints. (2) Manual Mesh Editing: User-guided manipulation of mesh geometries and topologies. (3) Point-Based Registration: Integration of point/curve/surface-based registration algorithms into the inverse-problems workflow.