Empirical Power Analysis Tool for fMRI

fMRI 经验功率分析工具

• 批准号: 10503064
• 负责人: Stephanie Noble
• 金额: $ 12.27万
• 依托单位: YALE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-07-15 至 2023-06-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10503064
• 关键词: Address; Algorithmic Analysis; Algorithmic Software; Algorithms; BRAIN initiative; Benchmarking; Brain; Characteristics; Complex; Computer software; Data; Data Set; Databases; Development; Dimensions; Disease; Documentation; Functional Magnetic Resonance Imaging; Funding; Human; Individual; International; Internet; Literature; Maps; Memory; Methodology; Methods; Modeling; Modernization; Online Systems; Participant; Patients; Pilot Projects; Procedures; Production; Professional Competence; Recommendation; Reproducibility of Results; Research; Research Design; Research Personnel; Sample Size; Sampling; Sampling Studies; Scanning; Specific qualifier value; Testing; To specify; Training; Work; base; career; design; imaging study; interactive tool; interest; power analysis; simulation; study characteristics; symposium; tool; web app; web-based tool

PROJECT SUMMARY Functional magnetic resonance imaging (fMRI) research has transformed our understanding of human brain function and disease and is flourishing under unprecedented international funding, including dedicated support from the BRAIN Initiative. However, recent work has exposed an endemic lack of statistical power (i.e., ability to detect effects of interest) in typical fMRI studies, leading to findings that do not replicate or uncover only a small tip of the iceberg of true effects. This arises in large part because performing proper power analyses to guide fMRI study design is not straightforward. First, it is difficult to estimate expected effects based on the literature, and study sample sizes are already so small that even smaller pilot data may not yield helpful estimates. Furthermore, fMRI data and inferential algorithms are complex, yet existing fMRI power analysis tools rely on relatively limited simulations, parametric estimates, and omit the most popular inferential procedures. As a result, fMRI researchers often perform misleading power analyses or avoid power analyses altogether, missing a critical opportunity to optimally design studies to detect desired effects. To address this gap, we will create a power analysis algorithm and tool tailored for standard fMRI studies that leverages: 1) large existing datasets to define typical study effects, and 2) recently developed methods for benchmarking power of complex inferential procedures. Finally, it will be designed to provide tailored recommendations and be easy to use, thus promoting its utility to everyday researchers. In Aim 1 (K99), we will create database of effect size maps for typical study designs using large, publicly available datasets and build a web app for exploring these maps. We will use this database in Aim 2 (R00) to design a post hoc power calculator algorithm to estimate power for typical study designs. Aim 3 (R00) will refine this algorithm by creating a meta-regression model that incorporates additional study and participant factors to provide a more tailored estimate of power for an individual researcher. Finally, in Aim 4 (R00) we will create and disseminate an easy-to use web-based tool for performing the “tailored” power analysis, notably only requiring the user to specify information readily available to them. This proposal will result in the first algorithm and tool to perform an empirical power analysis for fMRI study planning, with a potential user base that includes all researchers planning an fMRI study using typical designs. This will enable researchers to more easily and accurately plan well-powered studies, thus promoting more robust and reproducible findings in the field. Furthermore, this proposal will provide training in production-ready web development, study aggregation methods, and independence-oriented professional competencies, which will facilitate my transition to an independent research career leading statistical methodology development in fMRI.

项目摘要 功能磁共振成像（fMRI）研究改变了我们对人脑的理解 功能和疾病，并在前所未有的国际资金下荧光，包括专用支持 来自大脑倡议。但是，最近的工作已经暴露了统计能力的地方性缺乏（即能够 在典型的fMRI研究中检测感兴趣的影响，导致发现不复制或仅发现小的发现 真正影响的冰山一角。这在很大程度上是因为执行适当的功率分析以指导 fMRI研究设计并不简单。首先，很难根据文献估算预期效应， 研究样本量已经很小，以至于甚至较小的先导数据可能无法产生有用的估计。 此外，fMRI数据和推论算法很复杂，但现有的fMRI功率分析工具依靠 相关的有限模拟，参数估计并忽略了最流行的推论程序。因此， FMRI研究人员经常进行非法分析或完全避免使用电源分析，缺少关键 最佳设计研究以检测所需效果的机会。为了解决这一差距，我们将创建一个力量 分析算法和针对利用标准fMRI研究量身定制的工具：1）现有的大型现有数据集定义 典型的研究效果，以及2）最近开发了基准测试复杂推理能力的方法 程序。最后，它将设计为提供量身定制的建议并易于使用，从而促进 它对每天研究人员的实用性。在AIM 1（K99）中，我们将创建典型研究效果尺寸图的数据库 使用大型公开数据集进行设计，并构建一个用于探索这些地图的Web应用程序。我们将使用这个 AIM 2（R00）中的数据库设计一个事后功率计算器算法以估算典型研究的功率 设计。 AIM 3（r00）将通过创建包含其他的元回归模型来完善该算法 研究和参与因素为单个研究人员提供更量身定制的权力估计。最后，在 AIM 4（R00）我们将创建并传播一种易于使用的基于Web的工具，以执行“量身定制”的功率 分析，特别是仅要求用户可以随时指定信息。该提议将导致 在第一种算法和工具中，用于对fMRI研究计划进行经验能力分析，并具有潜力 用户群包括所有研究人员使用典型设计进行fMRI研究的研究人员。这将使研究人员 更轻松，更准确地计划了众多的研究，从而促进了更健壮和可重现的发现 在现场。此外，该提案将提供有关生产的Web开发培训，研究 汇总方法和独立的专业能力，这将有助于我的过渡 在fMRI的独立研究职业领先的统计方法发展。