Computational Modeling of Anatomical Shape Distributions

描述（由申请人提供）：从医学图像中分割出详细的、患者特定的模型，可以为手术计划和导航提供宝贵的帮助。目前的分割方法在面对细微的强度边界时，往往会出现错误。增加对结构的预期形状和形状正常变化范围的了解，可以通过将其引导到最可能与图像信息一致的形状来极大地改进分割。由此产生的分割可用于计划外科手术，并且当登记给患者时，可以在关键结构周围提供导航指导。许多神经系统疾病，如阿尔茨海默氏症、精神分裂症和胎儿生长受限，都会影响特定解剖区域的形状。为了了解这些疾病的发展和进展，以及开发将实例分为患病类或正常类的方法，1需要捕获种群之间形状分布差异的方法。我们的目标是开发和验证从图像中学习解剖形状及其可变性的简明表示的方法，建模形状分布将通过偏向更可能的形状来改进分割算法。它还将使人口研究中基于形状的定量分析成为可能，在这些研究中，成像被用于研究人口之间的解剖差异，以及人口内部的变化，例如年龄的变化。拟议的研究建立在先前的分割和形状分析方法的基础上，使用计算机视觉和机器学习工具应用于形状表示，基于形状的分割和人口研究的形状分析问题。我们计划进一步发展这些方法，并与我们的合作者在几个不同的应用中验证它们，包括手术计划、新生儿成像和基于图像的衰老和阿尔茨海默病研究。

DESCRIPTION (provided by applicant): Segmentation of detailed, patient-specific models from medical imagery can provide invaluable assistance for surgical planning and navigation. Current segmentation methods often make errors when confronted with subtle intensity boundaries. Adding knowledge of expected shape of a structure, and the range of normal variations in shape, can greatly improve segmentation, by guiding it towards the most likely shape consistent with the image information. The resulting segmentations can be used to plan surgical procedures, and when registered to the patient, can provide navigational guidance around critical structures. Many neurological diseases, such as Alzheimer's, schizophrenia, and Fetal Growth Restriction, affect the shape of specific anatomical areas. To understand the development and progression of these diseases, as well as to develop methods for classifying instances into diseased or normal classes, 1 needs methods that capture differences in shape distributions between populations. Our goal is to develop and validate methods for learning from images concise representations of anatomical shape and its variability, Modeling shape distributions will improve segmentation algorithms by biasing the search towards more likely shapes. It will also enable quantitative analysis based on shape in population studies, where imaging is used to study differences in anatomy between populations, as well as changes within a population, for example with age. The proposed research builds on prior methods for segmentation and shape analysis, using tools from computer vision and machine learning applied to questions of shape representation, shape based segmentation and shape analysis for population studies. We plan to further develop the methods and to validate them with our collaborators in several different applications, including surgical planning, neonatal imaging and image-based studies of aging and Alzheimer's disease.