Bayesian Spatial Point Process Modeling of Neuroimage Data

描述(申请人提供)：功能神经成像已成为非侵入性研究正常人群和临床人群大脑的基本工具。在过去的20年里，使用神经成像方法的研究量一直在急剧增长。这种研究的爆炸式增长得到了一种简单且计算效率高的方法的支持，这种方法被称为质量单变量方法(MUA)。然而，尽管它被普遍使用，但MUA有几个局限性：1)无法推断效应的确切位置；2) 不能正确地解释受试者之间的空间异质性和效应的空间结构；3)它不是为点模式数据而设计的，例如来自神经成像荟萃分析研究的数据，也不是来自多发性硬化症病变的二进制价值数据。为了克服MUA的这些局限性，我们建议开发贝叶斯统计模型，明确解决这些问题。特别是，我们将开发分层贝叶斯空间点过程模型来分析神经成像坐标水平数据(例如，当只有激活中心的峰值位置可用时，例如在神经成像元分析数据中的情况)、二进制成像数据(例如从多发性硬化症病变获得的数据)和神经图像体素水平数据的分层贝叶斯空间过程/空间点过程模型(例如，当在一组受试者上获得整个对比度或t统计图像时)。更多 最近，神经学家一直在收集纵向数据和横断面数据，目的是研究疾病的进展。关于纵向神经影像数据的分析工作还很少，因此我们进一步建议扩展我们的建模，以在数据中包含纵向方面以及横截面方面。我们将实施和优化我们的方法，并将软件向公众开放。这项工作的一个显著特点是，这些模型可以用来帮助预测/诊断神经精神/神经退行性疾病和障碍。因此，我们的模型将有助于理解神经精神和神经退行性疾病的发展，以及目前的方法/模型无法回答的正常大脑发育。反过来，这将有助于我们了解人类大脑在正常和疾病状态下的情况。

DESCRIPTION (provided by applicant): Functional neuroimaging has become an essential tool for non-invasively studying the brain of normal and clinical populations. The volume of research using neuroimaging methods has been growing dramatically in the last 20 years. This explosion of research has been supported by a simple and computationally efficient method known as the mass univariate approach (MUA). Despite its common use, however, there are several limitation to the MUA: 1) the inability to infer on the exact location of an effect; 2) the inability to properly account for spatial heterogeneity amongst subjects and the spatial structure of the effect; and 3) it is not designed for point pattern data, such as that from a neuroimaging meta analysis study, nor the binary valued data from multiple sclerosis lesions. To overcome these limitation of the MUA, we are proposing the development of Bayesian statistical models that explicitly address these issues. Specially, we will develop hierarchical Bayesian spatial point process models to analyze neuroimaging coordinate-level data (e.g. when only the peak location of the activation centers are available such as is the case in neuroimaging meta analysis data), binary imaging data (such as that obtained from multiple sclerosis lesions) and hierarchical Bayesian spatial process/spatial point process models for neuroimage voxel-level data (e.g. when the entire contrast or t-statistic image is available on a group of subjects). More recently, neuroscientists have been collecting longitudinal data, as well as cross-sectional data, with the intent of studying progression of disease. There is little work done on the analysis of longitudinal neuroimaging data, so we further propose to extend our modeling to incorporate the longitudinal aspect in the data as well as the cross-sectional aspect. We will implement and optimize our methods and make the software available to the public. One notable feature of this work is that the models can be used to help predict/diagnose neuropsychiatric/neurodegenerative diseases and disorders. Thus our models will assist in understanding the development of neuropsychiatric and neurodegenerative disorders, as well as normal brain development, that cannot be answered by current methods/models. This, in turn, will aid in our understanding of the human brain in normal and diseased states.