Clustering Patterns of Structural and Functional Neuroimaging Markers to Examine Individual Variability in Healthy Populations.

在神经影像学和认知神经科学文献中，越来越多的人认识到个体差异的重要性：总体或平均的群体反应不能很好地映射到个体反应。这种个体差异已被证明是描述广泛的基本人脑功能的实质性障碍。在精神病学和其他临床研究中，变异性一直是一个特别的挑战，在这些研究中，由于对一般生物异质性的描述不佳，对疾病生物标志物的寻找受到了阻碍。我们研究计划的目标是确定稳定、可靠和表型上有用的健康个体群体，这些个体具有相似的大脑功能和结构特征。这将增强我们对一般人群变异性的理解，并验证临床样本后续翻译的工具。我们建议通过应用和开发先进的数据驱动方法来识别在健康对照人群中具有相似大脑结构和功能模式的个体的子群体或“集群”来做到这一点。我们将使用大型多模态神经成像数据集（包括公开可用的和本地生成的）并检查大脑中的功能和结构连接。我们将比较一系列统计方法来识别子群体（例如，分层聚类，k-均值聚类，光谱聚类和相似网络融合）。聚类解决方案的稳定性/可靠性将通过排列引导方法进行评估，在这种方法中，数据将被重复采样，并评估聚类分配的稳定性。然后，我们将使用多元偏最小二乘方法询问跨子组的脑-行为关系，这将允许识别整个样本中常见的脑-行为关系或特定个体集群的独特关系。这项工作有可能促进我们对普通人群中大脑结构和功能的异质性的理解，以及这与行为的关系。这将作为今后可应用于临床人群的方法模板。一旦我们建立了可靠的方法来识别重要的个体亚群，或者个体在维度上不同的神经影像学指标，这些方法就可以转移到临床样本中，并作为临床人群中数据驱动的生物分型工具进行重新评估。

There is a growing appreciation of the importance of individual variability in neuroimaging and cognitive neuroscience literature: aggregated or averaged group responses do not map well on to individual responses. This Individual variability has proven a substantive obstacle to describing a wide range of basic human brain function. Variability has been a particular challenge in psychiatric and other clinical research where the search for biological markers of disease has been hampered by poor characterization of general biological heterogeneity. The objective of our research program is to identify stable, reliable, and phenotypically useful groups of healthy individuals with similar functional and structural brain characteristics. This will enhance our understanding of variability in the general population, as well as validate tools for later translation in clinical samples. We propose to do this by applying and developing advanced data driven approaches to identify sub-groups or `clusters' of individuals with similar patterns of brain structure and function, within a healthy control population. We will use large multi-modal neuroimaging data sets (both publicly available and locally generated) and examine functional and structural connectivity in the brain. We will compare a range of statistical approaches to identify sub-groups (e.g. hierarchical clustering, k-means clustering, spectral clustering, and similarity network fusion). The stability/reliability of clustering solutions will be assessed via a permuted bootstrapping approach, in which the data will be repeated subsampled and the stability of cluster assignments assessed. We will then interrogate brain-behavior relationship across sub-groups using a multivariate partial least squares approach, which will allow for the identification of brain-behavior relationships which are common across entire samples or unique to specific clusters of individuals. The proposed work has the potential to advance our understanding heterogeneity of brain structure and function within the general population, and how this related to behavior. This will serve as a template for approaches that can later be applied to clinical populations. Once we have established reliable approaches for identifying important sub-groups of individuals, or neuroimaging metrics along which individuals vary dimensionally, these approaches can be transitioned into clinical samples and re-evaluated as tools for data-driven biotyping in clinical populations.