Diffeomorphic Statistical Shape Models and Damage Quantification in Psoriatic Arthritis

银屑病关节炎的微分形统计形状模型和损伤​​量化

• 批准号: 2436904
• 金额: --
• 依托单位: University of Bath
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2020
• 资助国家: 英国
• 起止时间: 2020 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2436904
• 关键词: Diffeomorphic; Statistical; Shape; Models; Damage

The study of shape and variations in shape, is an exceedingly important part of computational anatomy and is useful in study of disease progression in various conditions; e.g. Psoriatic Arthritis. Standard Statistical Shape models in the computer vision literature do not preserve topology or regularity of the underlying family of contours, especially for large variations in shape. This is a drawback in medical applications where accurate models of shape and variations in shape, are an important tool in disease stratification.The proposed research project will develop Diffeomorphic Statistical Shape models. This uses ideas from the Large Deformation Diffeomorphic Metric Mapping (LDDMM) theory. This theory is typically used in nonrigid registration, computing minimal energy diffeomorphisms aligning two geometric structures of interest. We will use LDDMM to model variations in shape via action of diffeomorphism on an initial template. This guarantees topological consistency, as well as regularity of the deformed template. Shapes can be represented as point clouds, contours or measures in the more general case. This representation has links to the optimal transport literature. We combine this with Gaussian processes, and Gaussian Process Latent Variable Models to construct a generative model for shape. We place Gaussian process priors, on time dependent velocity fields that generate diffeomorphic flows. This prior will be conditioned on an observed shape family, to fit variations in shape to variations in fields. To regularise the shape generation process, we will extend this initial model using conservation properties and PDE arising from the Hamiltonian formulation of the LDDMM theory. In particular, momentum and energy conservation energy properties can be used to place GP priors over time dependent momentum and initial momentum. Our primary application of the proposed shape models will be to medical imaging and diagnosis. We will work with a dataset of shapes of hand bones with Psoriatic Arthritis or PsA (collected by the NHS). The shapes are in the form of landmarked X-Rays, annotated and scored by an expert clinician. Our generative shape model, will be used to model shape and shape variations in bones with PsA in a topologically consistent manner. In the long term, we aim to build a probablistic scoring tool for damage in PsA using the labelled dataset as training data.Upon success, such a tool will improve the diagnostic process in PsA and result in better understanding of the relationships between shape variation and disease trajectories in PsA. Healthcare technologies that 'optimise patient-specific illness prediction, accurate diagnosis and effective intervention' is one of the key themes in recent research supported by EPSRC. Therefore, this research is highly relevant to the research council.

形状和形状变化的研究是计算解剖学的一个非常重要的部分，并且在各种疾病的疾病进展研究中是有用的;例如银屑病关节炎。计算机视觉文献中的标准统计形状模型不保留底层轮廓族的拓扑或规则性，特别是对于形状的大变化。这是一个缺点，在医疗应用中，准确的形状模型和形状的变化，是一个重要的工具，在疾病分层。拟议的研究项目将开发同构统计形状模型。这使用来自大变形纯度量映射（LDDMM）理论的思想。该理论通常用于非刚性配准，计算对准两个感兴趣的几何结构的最小能量微分同胚。我们将使用LDDMM通过初始模板上的仿射动作来模拟形状的变化。这保证了拓扑的一致性，以及变形模板的规则性。在更一般的情况下，可以将点云、轮廓或测量表示为点云。这种表示与最优运输文献有联系。我们联合收割机，这与高斯过程，高斯过程潜变量模型构建一个生成模型的形状。我们把高斯过程先验，时间依赖的速度场，产生的非纯流。该先验将以观察到的形状族为条件，以使形状的变化适应场的变化。为了规范化的形状生成过程中，我们将扩展这个初始模型使用守恒性质和PDE所产生的哈密尔顿制定的LDDMM理论。特别地，动量和能量守恒能量属性可以用于放置GP先验随时间的依赖动量和初始动量。我们提出的形状模型的主要应用将是医学成像和诊断。我们将使用银屑病关节炎或PsA手骨形状的数据集（由NHS收集）。这些形状是由专家临床医生注释和评分的标志性X射线的形式。我们的生成形状模型，将被用来模拟形状和形状变化的骨骼与PsA在拓扑一致的方式。从长远来看，我们的目标是使用标记数据集作为训练数据建立PsA损伤的概率评分工具。成功后，这样的工具将改善PsA的诊断过程，并更好地了解PsA形状变化和疾病轨迹之间的关系。“优化患者特定疾病预测，准确诊断和有效干预”的医疗保健技术是EPSRC最近支持的研究的关键主题之一。因此，这项研究与研究理事会高度相关。