Data mining in large functional neuroimaging databases

功能性核磁共振成像提供了一种间接测量大脑活动的方法，具有良好的时间和空间分辨率。这些大量的信息需要被总结以便于解释。聚类分析可以识别出具有休息时相关活动的空间分布区域，称为休息状态网络（RSNs）。我的研究项目的首要目标是捕捉这些rsn之间的连接如何在阿尔茨海默病的反应中发生变化。本提案的短期目标针对长期目标带来的技术挑战：(1)rsn的数量（和大小）是研究者的任意选择，即大脑可以分解为大型分布式系统，也可以分解为专门的大脑区域。我想确定哪种分解程度最适合用来检测阿尔茨海默病对大脑连接的影响。(2)目前，rsn是在一个群体的水平上推导出来的，所有被试的大脑都是根据其形态排列后得出的。在这个过程之后，大脑的功能单位仍然可以在个体之间呈现差异。我们将开发新的聚类技术，以匹配跨主题的剩余功能变异性。这项手术有望更好地揭示阿尔茨海默病对大脑连通性的有害影响，这可能被受试者之间的功能错误登记所掩盖；(3) rsn和相关连通性度量的生成是一个涉及许多步骤和决策的长期过程。我们将使用真实和模拟数据集，系统地测试fMRI数据集预处理中做出的各种选择对目标1和2结果的影响。我预计在这个研究项目中取得的进展将大大提高基于静息状态fMRI结果测量的可靠性和敏感性，这对临床和基础神经科学都具有重要意义。

Functional MRI gives an indirect measure of brain activity with a good temporal and spatial resolution. This massive amount of information needs to be summarized to be interpretable. Cluster analysis can identify spatially distributed regions with a temporally correlated activity at rest, called resting-state networks (RSNs). The overarching goal of my research program is to capture how the connectivity between these RSNs changes in response to Alzheimer's disease. The short-term objectives of this proposal target the technical challenges raised by this long-term goal: (1) The number (and size) of RSNs is an arbitrary choice of the investigator, i.e. the brain can be decomposed into large distributed systems as well as into specialized brain regions. I want to determine which level of decomposition is the most optimal to detect the effects of Alzheimer's disease on brain connectivity. (2) Currently, RSNs are derived at the level of a group after the brains of all subjects have been aligned based on their morphology. The functional units of the brain can still present variations amongst individuals after this process. We will develop novel clustering techniques that will match this remaining functional variability across subjects. This operation is expected to reveal better the deleterious effects of Alzheimer's disease on brain connectivity, that could have been masked by the functional misregistration between subjects; (3) The generation of RSNs and associated connectivity measures is a long process involving many steps and decisions. We will systematically test the impact of various choices made in the preprocessing of fMRI datasets on the outcomes of Objectives 1 & 2, using both real and simulated datasets. I anticipate that the developments made in this research program will substantially increase the reliability and sensivity of outcome measures based on resting-state fMRI with important implications for both clinical and basic neuroscience.