Characterization of Neurodevelopmental Disease Trajectories using Richly Annotated Sequences of Graphs (RICHGRAPH)

在欧洲，精神障碍是自杀、残疾和提前退休的主要原因。虽然磁共振成像的最新发展为大脑发育打开了前所未有的窗口，但计算方法分析手头纵向和多模态成像信息的全部潜力尚未得到充分利用。在精神病学神经成像研究长期专注于大脑的特定区域（局部分析）之后，新兴的连接组学领域通过建立图作为大脑的表征，使拓扑特性的表征（全局分析）成为可能。然而，目前的技术要求将可用的多模态信息简化为一组简单的边缘，这些边缘在单个权重中捕获不同大脑区域之间的关系，因此无法利用多种mri衍生的定量参数作为潜在的互补信息源。他们也无法模拟网络动力学。受社交网络分析领域重大进展的启发，该项目将开发基于丰富图形的下一代连接组学技术，从而能够对来自多个来源的异构数据进行整体处理。应用于富图序列的先进机器学习技术将允许基于局部组织特征和全局网络特征分析和预测神经发育轨迹。包括纵向成像、遗传、环境和行为数据在内的大量患者和对照组的综合验证研究将进行，其长期目标是：(1)建立新的多模态和纵向成像生物标志物，代表临床症状出现之前的病理变化；(2)将脑成像特征与基因变异联系起来；(3)促进我们对潜在生物学过程的理解，为开发新的治疗方法铺平道路。

Mental disorders are the main cause of suicide, disability, and early retirement in Europe. While recent developments in magnetic resonance imaging have opened unprecedented windows into the developing brain, the full potential of computational methods to analyse the longitudinal and multi-modal imaging information at hand has not yet been exploited. After psychiatric neuroimaging research has long been focusing on dedicated regions in the brain (local analysis), the emerging field of connectomics has enabled a characterization of topological properties (global analysis) by establishing graphs as a representation of the brain. Current techniques, however, require the available multi-modal information to be reduced to a simple set of edges that capture the relationship between different brain regions in a single weight and thus fail to exploit the manifold MRI-derived quantitative parameters as potentially complementary sources of information. They are also unable to model network dynamics. Inspired by major progress in the field of social network analysis, this project will develop the next generation of connectomics techniques based on rich graphs that enable holistic processing of heterogeneous data from multiple sources over time. Advanced machine learning techniques applied to sequences of rich graphs will allow analyses and prediction of neurodevelopmental trajectories based on both local tissue characteristics and global network features. Comprehensive validation studies with large cohorts of patients and controls that include longitudinal imaging, genetic, environmental, and behavioral data will be performed with the long-term goal of (1) establishing novel multi-modal and longitudinal imaging biomarkers that represent pathological changes before the appearance of clinical symptoms, (2) linking brain imaging traits to gene variants and (3) advancing our understanding of the underlying biological processes to pave the way for development of new treatments.