Parcellating Infant Cerebral Cortex based on Developmental Patterns of Multimodal MRI

项目摘要 大规模纵向多模式婴儿脑MRI数据集的可用性不断增加，例如Baby Connectome Project(BCP)为精确绘制动态轨迹提供了前所未有的机会 这对于理解正常生长和神经发育障碍是至关重要的。一个 主要障碍是严重缺乏基于皮质表面的计算工具、地图集和地块 对具有挑战性的婴儿磁共振成像的分析，典型的表现为低组织对比度和区域异质性， 大脑皮层特性的动态变化。为了填补这一空白，我们率先推出了一套全面的婴儿- 专用皮质表面分析工具和地图集。我们关于早期大脑发育的工具和发现 已经在NIMH的2015-2020年战略计划中得到了强调。然而，计算方法仍然是 根据纵向多模式磁共振成像的动态脑特性，婴儿缺乏皮质分离。 在广泛的婴儿神经成像应用中，分割是一个先决条件，例如，区域定位、间隔区定位、 个体变异性调查、研究间比较、统计敏感性提升、节点定义 网络分析和特征约简，用于识别大脑疾病。因此，本项目的重点是 为人口水平和个性化的婴儿创建和传播新的计算工具 利用多种互补大脑特性的发育模式进行皮质分割，以及 将它们应用于更好地理解个体间的差异和早期大脑发育。这个 动机是多种属性(例如，皮质厚度、折叠性、扩散性、 髓鞘含量、表面积、结构和功能的连通性)在本质上反映了婴儿 底层微结构及其连通性的变化，共同决定了 每个地区。因此，发展模式是在发展、微观结构、 用于早期大脑发育研究的功能和连接性。为了实现这一目标，我们提出了四个具体的建议 目标。在目标1中，我们将开发一种新的种群级别的皮质分割方法，基于 异质多峰信息非线性融合的多属性发展模式 从一大群婴儿中分离出来。在目标2中，我们进一步提出了一种新的个性化分类方法 根据每个婴儿自己的多模式发育模式对每个婴儿的皮质表面进行分析，从而解释 显著的学科间变异性。我们将利用人口层面的划分来引导个性化 通过图形切割以迭代的方式进行分割，从而产生精确的个性化分割 在不同个体之间很容易进行比较。在目标3中，了解每一个个体之间显著的差异 分割的区域，我们将发现每个皮质属性的代表性区域外观模式 根据大脑皮质特性的多尺度空间-频率特征，从大量婴儿人群中提取数据 通过球面小波进行映射。在目标4中，利用我们的工具、地图集和地块，我们将绘制 每个物业的每种典型模式的多模式发展轨迹，并调查其 与行为/认知得分的关系。最后，我们将自由地释放我们的工具、包裹和 向公众公布经过处理的BCP数据。

Project Abstract The increasing availability of large-scale longitudinal multimodal infant brain MRI datasets, e.g., the Baby Connectome Project (BCP), provides an unprecedented opportunity to precisely chart the dynamic trajectories of early brain development, essential for understanding normative growth and neurodevelopmental disorders. A major barrier is the critical lack of computational tools, atlases and parcellations for cortical surface-based analysis of the challenging infant MRI, which typically exhibits low tissue contrast and regionally-heterogeneous, dynamic changes of cortical properties. To fill this gap, we have pioneered a comprehensive set of infant- dedicated cortical surface analysis tools and atlases. Our tools and discoveries on early brain development have been highlighted in NIMH’s 2015-2020 Strategic Plan. However, computational approaches are still lacking for infant cortical parcellation based on the dynamic brain properties from longituidnal multimodal MRI. Parcellation is a prerequsite in a wide variety of infant neuroimaging applications, e.g., region localization, inter- individual variability investigation, inter-study comparison, statistical sensitivity boosting, node definition for network analysis, and feature reduction for identificaiton of brain disorders. Hence, this project is focused on creating and disseminating novel computational tools for both population-level and individualized infant cortical parcellation utilizing developmental patterns of multiple complementary brain properties, and applying them to better understanding of inter-individual variability and early brain development. The motivation is that the dynamic development of multiple properties (e.g., cortical thickness, folding, diffusivity, myelin content, surface area, structural and functional connectivity) in infants essentially reflects the rapid changes of underlying microstructures and their connectivity, which jointly determine the functional principle of each region. Hence, developmental patterns are ideal for deriving distinct regions in development, microstructure, function, and connectivity for early brain development studies. To achieve this goal, we propose four specific aims. In Aim 1, we will develop a novel method for population-level cortical parcellation based on developmental patterns of multiple properties, by nonlinear fusion of heterogeneous multimodal information from a large population of infants. In Aim 2, we further propose a novel approach for individualized parcellation of each infant’s cortical surfaces based on its own multimodal developmental patterns, thus accounting for remarkable inter-subject variability. We will leverage the population-level parcellation to guide the individualized parcellation in an iterative manner via graph cuts, thus leading to precise individualized parcellations that are easily comparable across individuals. In Aim 3, to understand the remarkable inter-individual variability in each parcellated region, we will discover the representative regional appearance patterns of each cortical property from a large infant population, based on multi-scale spatial-frequency characterizations of cortical property maps via spherical wavelets. In Aim 4, leveraging our tools, atlases, and parcellations, we will chart the multimodal developmental trajectories for each representative pattern of each property and investigate their relationships with behavioral/cognitive scores. Finally, we will freely release our tools, parcellations and the processed BCP data to the public.