Pattern Analysis of fMRI via machine learning/sparse models: application to brain development

摘要 静息状态功能磁共振成像(RsfMRI)提供了可重复性的、与任务无关的相干生物标志物 连接不同大脑区域的功能活动。被提议的项目的主要目标是利用 信号处理和机器学习方法在提取临床有用生物标志物方面的进展 基于功能连接的模式，并在一项大型大脑研究中测试这些生物标记物 发展。我们方法论的核心是1)计算特定于主题的功能分区 定义节点，用于表征个性化的功能大脑网络；2) 提取稀疏连接模式用于稳健表示脑网络；3)捕获 在特定人群中跨个体的大脑网络的异质性；以及4)派生个性化 在一项关于大脑发育的大型研究中，大脑连接对精神病风险的预测指标。这 将开发一套新的功能连通性分析工具，并基于数据进行验证 来自人类连接组项目和费城神经发育队列(PNC)。 最后，将这些技术应用于PNC数据，以描绘网络中的异构性 有精神病谱系症状的青年的发展。我们的假设是， 患有精神病谱系症状的青少年的功能连接将不同于那些 在典型的发育青少年中，这种差异将显示出高度的异质性， 与病理性神经发育轨迹的潜在异质性有关。此外，我们 期望机器学习技术将允许我们在个人的基础上预测 有精神病症状的青少年将保持稳定，并将恢复正常，以及 根据他们的基线功能连接特征，这将进展为精神病。我们的 方法一般适用于rsfmri的时空检测和量化研究。 不同领域的功能连接模式，包括诊断脑部异常 神经精神疾病，以及发现功能连接与不同认知之间的关联 功能。所有方法都将公之于众，并成为 更广泛的神经科学界。

Abstract Resting state fMRI (rsfMRI) provides reproducible, task-independent biomarkers of coherent functional activity linking different brain regions. The main goal of the proposed project is to leverage advances in signal processing and machine learning methods to derive clinically useful biomarkers based on patterns of functional connectivity, and to test these biomarkers in a large study of brain development. Central to our methodology are 1) computing a subject-specific functional parcellation of the brain, which defines nodes for characterizing individualized functional brain networks; 2) extracting sparse connectivity patterns for robustly representing brain networks; 3) capturing heterogeneity in brain networks across individuals in a given population; and 4) deriving individualized predictive indices of psychosis risk from brain connectivity in a large study of brain development. This novel suite of functional connectivity analysis tools will be developed and validated based on data from the Human Connectome Project and the Philadelphia Neurodevelopmental Cohort (PNC). Finally, these techniques will be applied to PNC data in order to delineate heterogeneity in network development in youth with psychosis-spectrum symptoms. Our hypothesis is that patterns of functional connectivity in adolescents with psychosis-spectrum symptoms will be different from those in typically developing adolescents, and this difference will display a high degree of heterogeneity that is linked to underlying heterogeneity in pathologic neurodevelopmental trajectories. Moreover, we expect that machine learning techniques will allow us to predict on an individual basis which adolescents with psychosis-spectrum symptoms will remain stable, which will revert to normal, and which will progress to psychosis, based on their baseline functional connectivity signatures. Our methods are generally applicable to rsfMRI studies for detecting and quantifying spatio-temporal functional connectivity patterns in diverse fields, including diagnosing brain abnormalities in neuropsychiatric diseases, and finding associations of functional connectivity with different cognitive functions. All methods will be made publicly available and form an important new resource for the broader neuroscience community.