Analytic Methods for Functional Neuroimaging Data

描述（由申请人提供）：功能神经成像技术，包括功能磁共振成像（fMRI）和正电子发射断层扫描（PET），是心理健康研究的强大的非侵入性工具。fMR的解析方法！PET数据对于确定可以解决的实质性研究问题和确保推论的有效性都是至关重要的。该项目旨在为fMRI和PET数据开发最先进的统计方法，这将对临床实践和研究产生重要的心理健康影响。长期目标是：1)协助医生为精神疾病患者做出治疗决定，这里主要关注精神分裂症和重度抑郁症；2)建立一个建模框架，用于描述与任务相关的大脑活动，该活动可以解释空间关联，例如，大脑区域之间复杂的神经生理联系。在对临床心理健康实践产生影响的努力中，一个目标是开发一种新的方法来预测个体对治疗的特定反应。具体来说，目标是基于治疗前扫描和其他患者特征，预测特定患者治疗后与任务相关的大脑活动模式，并预测治疗后的最终症状反应。计划中的发展需要为精神分裂症患者和从未治疗过的抑郁症患者构建和验证贝叶斯层次模型和准确的分类算法。第二个目标是构建一个贝叶斯层次模型，用于推断大脑活动中与任务相关的变化，考虑不同空间位置（体素）之间的功能关联。这种方法将产生局部估计，类似于通常应用的方法，但将估计和调整关键的功能联系。通过建立基于非常适合数据的假设的模型，所提出的程序的一个主要优点是能够从相关体素中提取局部推断，通常会产生更准确的结果。第二个优点是，关于扩展解剖区域的测试可以包含体素间相关性的估计。目标2的空间建模发展可能会导致所提议的预测框架的扩展（目标1）。预测算法的成功开发将提供自然转化为临床环境的结果，以帮助告知医生关于精神病治疗的决定。此外，所提出的空间建模框架将对现有的功能神经成像数据分析方法做出新的贡献。本文对与精神分裂症、抑郁症和可卡因依赖相关的fMRI和PET数据的关注说明了所提出的方法在一系列精神健康障碍中的潜在适用性和相关性。

DESCRIPTION (provided by applicant): Functional neuroimaging technologies, including functional magnetic resonance imaging (fMRI) and positron emission tomography (PET), are powerful noninvasive tools for mental health research. Analytic methods for fMR! and PET data are critical both for determining substantive research questions that can be addressed and for ensuring the validity of inferences. This project seeks to develop state-of-the-art statistical methodology for fMRI and PET data that will have important mental health implications for clinical practice and research. The long-term goals are 1) to assist physicians in making treatment decisions for patients with psychiatric disorders, focusing here on schizophrenia and major depression and 2) to establish a modeling framework for characterizing task-related brain activity that accounts for spatial associations arising, for example, from complex neurophysiological links between brain regions. In an effort to make an impact on clinical mental health practices, one aim is to develop a novel approach to predict individual-specific responses to treatment. Specifically, the goal is to predict post- treatment patterns of task-related brain activity for a particular patient based on pre-treatment scans and other patient characteristics and to predict eventual symptom response to treatment. The planned developments entail constructing and validating a Bayesian hierarchical model and an accurate classification algorithm for schizophrenia patients and for never-treated depressed subjects. A second aim is to construct a Bayesian hierarchical model for making inferences regarding task-related changes in brain activity, accounting for functional associations between different spatial locations (voxels). This approach would yield localized estimates, similar to commonly applied methods, but would estimate and adjust for key functional linkages. By building a model based on assumptions that are well-suited to the data, a major advantage of the proposed procedure is the ability to draw localized inferences that borrow strength from related voxels, often yielding more accurate results. A second advantage is that tests about extended anatomical regions can incorporate estimates of between-voxel correlations. Spatial modeling developments from Aim 2 may give rise to extensions to the proposed prediction framework (Aim 1). Successful development of the predictive algorithms would provide results that translate naturally to a clinical setting to help inform physicians' decisions regarding psychiatric treatments. Furthermore, the proposed spatial modeling framework would be a novel contribution to existing analytic methods for functional neuroimaging data. The focus here on fMRI and PET data related to schizophrenia, depression, and cocaine-dependence illustrates the potential applicability and relevance of the proposed methods across a range of mental health disorders.