Improving the Detection of Activation in High Resolution fMRI using Multivariate

项目描述（由申请人提供）：本项目总体目标是开发一个用于fMRI数据分析的本地多元分析软件包。它将为心理学家和神经科学家提供一个更强大的工具，用先进的多变量方法分析他们的功能磁共振成像数据。这个项目将带来更好的大脑激活图，从而促进对大脑功能目前未知方面的发现。质量-单变量分析，如一般线性模型（GLM），是目前流行的功能磁共振成像数据分析方法。然而，由于常规应用固定各向同性空间高斯平滑，存在激活边缘模糊和弱激活区域检测电位消除的问题。局部多变量方法，如典型相关分析（CCA）及其变体，已被证明可以显著提高fMRI激活的检测能力，并改善激活图。作为一种优势，CCA使用自适应空间滤波核，可以在噪声环境中更好地准确提取信号。然而，该方法存在一些缺点，特别是空间特异性低、计算时间长、单因素实验设计受限。此外，还没有一种参数估计方法来确定家族错误率，没有对群体分析的扩展进行研究，也没有使用核方法将fMRI数据的局部CCA扩展到非线性CCA的研究。这些缺陷阻碍了局部CCA方法在fMRI神经科学研究中的广泛应用。在本提案中，我们的目标是使用新颖的局部多变量分析方法（基于CCA）来消除这些缺点，并开发一个软件工具来扩大其在神经科学研究界的广泛应用。我们希望这个软件工具对神经科学研究特别有价值，在神经科学研究中，需要检测弱激活或空间局部激活模式。随着高分辨率成像和计算机能力的进步，我们预计对这种软件工具的需求会增加，从而推动大脑功能的新发现和更精确的激活空间定位。作为一个特别的

DESCRIPTION (provided by applicant): The overall goal of this project is to develop a local multivariate analysis software package for fMRI data analysis. It will provide psychologists and neuroscientists a more powerful tool to analyze their fMRI data using advanced multivariate methods. This project will lead to better brain activation maps and thus promote the discovery of currently unknown aspects of brain function. Mass-univariate analysis, such as the general linear model (GLM), is the prevailing fMRI data analysis method. However, it suffers from blurring of edges of activation and potential elimination of the detection of weak activated regions due to routinely applied fixed isotropic spatial Gaussian smoothing. Local multivariate methods such as canonical correlation analysis (CCA) and its variants have been shown to significantly increase the detection power of fMRI activations and improve activation maps. As an advantage, CCA uses adaptive spatial filtering kernels to accurately extract the signal better in a noisy environment. However, there are several drawbacks, particularly low spatial specificity, long computational time, and single-factor experimental design limitation. Furthermore, a parametric estimation method does not exist to determine the family-wise error rate, no extension to group analysis has been investigated, and no studies extending local CCA to nonlinear CCA for fMRI data using kernel methods have been systematically carried out. All these drawbacks prevent local CCA methods from being widely accepted in neuroscience research in fMRI. In this proposal, our goals are to eliminate these drawbacks using novel local multivariate analysis methods (based on CCA) and to develop a software tool to widen its broader application in the neuroscience research community. We expect this software tool to be particularly valuable for neuroscience research where detections of weak activations or spatially localized patterns of activations are desired. As high resolution imaging and computer power advance, we expect an increase in demand for this software tool, thus advancing new discoveries of brain function and more precise spatial localization of activations. As a particular application, we will focus on studying memory actions using a novel event-related recognition paradigm to investigate the effects of familiarity and recollection in subregions of the medial temporal lobes (MTL) for high resolution fMRI. This research will advance our understanding of hippocampal/MTL contributions to memory, which can substantially advance our understanding of the memory deficits associated with a number of debilitating neurological and psychiatric conditions that show abnormalities in these regions, including mild cognitive impairment (MCI), Alzheimer¿s disease, schizophrenia, and major depression. More generally, it will provide psychologists and neuroscientists a more powerful tool to analyze their fMRI data using advanced multivariate methods.