Magnetic resonance imaging (MRI) is a technique that scans the anatomical structure of the brain, whereas functional magnetic resonance imaging (fMRI) uses the same basic principles of atomic physics as MRI scans but image metabolic function. A major goal of MRI and fMRI study is to precisely delineate various types of tissues, anatomical structure, pathologies, and detect the brain regions that react to outer stimuli (e.g., viewing an image). As a key feature of these MRI-based neuroimaging data, voxels (cubic pixels of the brain volume) are highly correlated. However, the associations between voxels are often overlooked in the statistical analysis. We adapt a recently proposed dimension reduction method called the envelope method to analyze neuoimaging data taking into account correlation among voxels. We refer to the modified procedure the envelope chain procedure. Because the envelope chain procedure has not been employed before, we demonstrate in simulations the empirical performance of estimator, and examine its sensitivity when our assumptions are violated. We use the estimator to analyze the MRI data from ADHD-200 study. Data analyses demonstrate that leveraging the correlations among voxels can significantly increase the efficiency of the regression analysis, thus achieving higher detection power with small sample sizes.

中文翻译：

---

在基于磁共振成像的研究中通过包络链进行有效估计

磁共振成像 (MRI) 是一种扫描大脑解剖结构的技术，而功能性磁共振成像 (fMRI) 使用与 MRI 扫描相同的原子物理学基本原理，但图像代谢功能。MRI 和 fMRI 研究的一个主要目标是精确描绘各种类型的组织、解剖结构、病理，并检测对外部刺激（例如，查看图像）做出反应的大脑区域。作为这些基于 MRI 的神经影像数据的一个关键特征，体素（脑体积的立方像素）高度相关。然而，体素之间的关联在统计分析中经常被忽视。我们采用最近提出的一种称为包络法的降维方法来分析神经成像数据，同时考虑到体素之间的相关性。我们将修改后的程序称为信封链程序。因为以前没有使用过包络链程序，我们在模拟中证明了估计器的经验性能，并在我们的假设被违反时检查它的敏感性。我们使用估计器来分析来自 ADHD-200 研究的 MRI 数据。数据分析表明，利用体素之间的相关性可以显着提高回归分析的效率，从而以小样本量实现更高的检测能力。