CRCNS: BrainPack: A Suite of Advanced Statistical Techniques for Mmulti-Subject, Multi-Group Neuroimaging Data Analysis

了解精神疾病的神经基础是发展有针对性的行为或药物治疗的关键组成部分。 鉴于脑成像设备的可用性和可获得性的快速进步，现在存在大量关于功能性神经成像结果的文献报道，这些结果将精神障碍（如精神分裂症）患者与健康人区分开来。然而，文献中充斥着复制失败的案例。成像作为精神疾病研究工具的普及以及结果缺乏一致性，提供了一个令人信服的证据，说明为什么要将资源投入到开发更可靠，更准确和更敏感的数据分析工具上。这些工具必须以健全的统计理论为基础，但也要适应数据现实带来的实际挑战。该项目开发了一套强大，灵敏和有效的统计方法，这将有助于神经科学家更好地了解精神疾病的病因。 这些工具灵敏度的提高也为评估新的治疗方法创造了更好的手段，因为它提供了对时间变化的更好评估，而这些变化由于其微妙的性质而目前难以捕捉。该项目开发的一套方法（BrainPack）是一个用于分析组级成像数据的综合系统，对测量信号的分布行为进行了最小的假设。它不需要一个先验模型的预期激活跨会话，可以有效地减少大数据集的大小包含大多数不相关的信息，占空间和时间的相关性，并量化组之间的差异和评估这些差异的统计意义。因此，BrainPack将为许多类型的神经影像学研究中常见的细微组差异提供强大的识别。这些进步是可推广的，并且容易适用于广泛的神经影像学研究和其他研究。

Understanding the neural bases of psychiatric disorders is a critical component in the development of targeted behavioral or drug therapies. Given the rapid advancements in availability of, and access to brain imaging equipment, there now exists a large literature reporting on functional neuroimaging results differentiating people with psychotic disorders (such as schizophrenia) from healthy people. The literature is littered, however, with failures to replicate. The popularity of imaging as a research tool in psychiatric disorders, and the lack of consistency in results, provide a compelling demonstration of why resources would be well invested on the development of more reliable, accurate and sensitive tools for analyzing data. These tools must be based on sound statistical theory, yet accommodate the actual, practical challenges caused by the realities of the data. The project develops a suite of robust, sensitive and effective statistical methods which will help neuroscientists better understand the etiology of psychiatric disorders. The enhanced sensitivity of these tools also creates a better means for evaluating new treatments, as it provides improved assessment of changes across time that are currently difficult to capture due to their subtle nature.The suite of methods developed in the project (BrainPack) is a comprehensive system for the analysis of group-level imaging data that makes minimal assumptions on the distributional behavior of the measured signal. It does not require an a priori model of the expected activation across sessions, can effectively reduce the size of large data sets containing mostly irrelevant information, account for spatial and temporal correlations, and quantify the discrepancy between groups and assess the statistical significance of these discrepancies. As such BrainPack will provide robust identification of subtle group differences that are common across many types of neuroimaging studies. These advancements are generalizable and readily adapted across a wide range of neuroimaging studies and beyond.