Statistical Methods for Neuroimaging Data

描述：(申请人提供)这项申请的长期目标是确立候选人作为多学科研究团队的领导成员和神经成像研究统计方法的独立研究员。培训计划是对 候选人S有很强的统计学背景，在神经科学、生物医学成像和心理健康方面都有基础。该研究计划侧重于精神健康的神经成像研究的新统计方法，特别是精神分裂症、焦虑症和成瘾。 其中一个目标是开发最先进的统计方法，以识别具有相似功能磁共振成像(FMRI)特征的大脑区域。彩色大脑图像的显示将区分集群，动态图像将描绘集群在时间、研究条件或距离指标上的变化。将对几种聚类方法进行比较和近似 将开发提高计算效率的方法。集群方法将允许评估受试者在执行任务、体验情绪状态或表现特定行为时使用多个大脑区域的情况。 其他具体目标包括开发正电子发射断层扫描(PET)和纳入受试者内部相关性的功能磁共振数据的统计模型。一种方法将使用具有相关误差和随机影响的线性模型。第二种方法将使用贝叶斯分层模型，直接考虑脑体素之间的空间相关性和时间相关性 通过各种协方差模型。另一种方法将通过体素特定参数的先验分布来直接建模时间相关性和间接地建模空间相关性，例如，PNOR导致相邻体素的相似性。此外，空间网络将被扩展到包括接近的体素。根据解剖学或生理学的联系。电脑 这些研究开发的软件将向神经成像科学家开放。

DESCRIPTION: (provided by applicant) The long-term objective of this application is to establish the candidate as a leading member of multi-disciplinary research teams and as an independent researcher in statistical methodology for neuroimaging studies. The training plan complements the candidate?s strong statistical background by providing foundations in neuroscience, biomedical imaging, and mental health. The research plan focuses on new statistical methods for neuroimaging studies of mental health, in particular, schizophrenia, anxiety disorder, and addiction. One aim is to develop state-of-the-art statistical methodology to identify brain regions exhibiting similar functional magnetic resonance imaging (fMRI) profiles. Displays of colored brain images will distinguish clusters, and dynamic images will depict cluster changes across times, study conditions, or distance metrics. Several clustering methods will be compared and approximate approaches will be developed to increase computational efficiency. The clustering methodology will allow evaluation of the use of multiple brain regions by subjects when performing tasks, experiencing emotional states, or exhibiting certain behaviors. Other specific aims include developing statistical models for positron emission tomography (PET) and fMRI data incorporating intra-subject correlation. One method will use linear models with correlated errors and random effects. A second method will use Bayesian hierarchical models directly accounting for spatial correlation between and temporal correlation within brain voxels through various covariance models. Another approach will model temporal correlation directly and spatial correlation indirectly through prior distributions of voxel-specific parameters, e.g., pnors inducing similarity for neighboring voxels. Also, spatial networks will be extended to include voxels that are ?close? according to anatomical or physiological connections. Computer software for these research developments will be made accessible to neuroimaging scientists.