Data-driven approach for identifying subgroups using fMRI connectivity maps

描述（由申请人提供）：希望了解人类大脑过程的功能磁共振成像（fMRI）研究人员越来越多地估计感兴趣区域（roi）之间的关系。总之，这些估计创造了一个大脑处理如何发生的“连接图”。对于大多数连接映射方法来说，一个普遍存在的问题是，为了获得可靠和有效的结果，它们需要个体之间的同质性。尽管有一致的证据表明，在对照和临床人群中，大脑过程在不同的人类样本中存在很大差异，但研究人员目前别无选择，只能依赖于同质性假设。因此，为了检验根据人口统计、行为或诊断指标产生的亚群体之间的差异，研究人员必须假设这些亚群体中的所有个体都是相同的。在神经成像领域需要数据驱动的方法，从他们的连接图中识别个体的亚组，以适应亚组内的异质性。数据驱动的亚组分类除了帮助研究人员了解亚组内的异质性外，还可以通过对整个样本进行亚组分类，识别与次优任务表现或特定诊断相关的大脑过程。本项目旨在通过开发一种分析功能磁共振成像数据的新方法来满足这一需求，该方法：1)得出可推广到总体的有效样本水平推断；2)确定个体的亚群分类；3)在个体水平上提供可靠的参数估计。在开发、验证和实施新程序后，由本文作者开发的一个成功的新算法构建的程序将免费向公众提供。

DESCRIPTION (provided by applicant): Functional MRI (fMRI) researchers wishing to understand human brain processes increasingly estimate relations among regions of interest (ROIs) across time. Together, these estimates create a "connectivity map" of how brain processing occurs. One ubiquitous issue for most connectivity mapping methods is that they require homogeneity across individuals for reliable and valid results to be obtained. Researchers currently have no choice but to rely on homogeneity assumptions despite consistent evidence suggesting that brain processes vary substantially across human samples within control and clinical populations. Thus to examine differences between subgroups created according to demographic, behavioral or diagnostic indices, researchers must assume that all individuals within these subgroups are the same. There is a need in the field of neuroimaging for data-driven methods for identifying subgroups of individuals from their connectivity maps to accommodate within-subgroup heterogeneity. Data-driven subgroup classification could identify brain processes which relate to suboptimal task performance or specific diagnoses by subgrouping the entire sample in addition to helping researchers understand heterogeneity within subgroups. The present project aims to fill this demand by developing a novel approach for analyzing fMRI data which: 1) arrives at valid sample-level inferences that may be generalized to the population; 2) identifies subgroup classification for individuals; and 3) provides reliable parameter estimates at the individual level. After developing, validating, and implementing the new procedure, a program which builds from a successful novel algorithm developed by the present authors will be made freely available to the public.