Longitudinal Mapping of Human Brain Development in the First Years of Life

人类生命最初几年大脑发育的纵向映射 摘要 该提案要求继续为位于查佩尔山的北卡罗来纳州大学的研究提供资金支持 开发量化人脑纵向结构变化的计算工具。前一 项目期间在推进老龄化纵向大脑分析的强大工具方面取得了极大的成功 个脑袋在这次更新中，我们寻求进一步推进全面纵向的强大计算工具， 在早期发育的大脑中的变化的表征。这符合我们的长期目标， 计算工具，用于纵向绘制整个人类生命周期的大脑进化图。能力成为 本项目开发的一种新的神经网络将允许对纵向体积数据和皮质骨数据进行统一的艾德和并行分析。 表面，促进在关键时期动态和空间异质结构变化的映射 大脑发育期。 该项目开发的工具将专门用于研究生命最初几年的人类大脑， 在结构和功能上都是动态发展的。我们将利用通过以下途径提供的MRI数据 婴儿连接组项目（BCP），涉及500名从出生到5岁的儿科受试者。的 BCP的结果将告知神经科学家正常健康的生长是什么样子的，并有助于发现 脑部疾病的早期表现。为了充分贝内这一独特的数据集， 婴儿MR图像通常表现出动态异质性， 随着时间的推移，然而，迄今为止开发的大多数计算工具主要集中在 成人受试者和不可靠时，适用于婴儿MRI。我们建议通过三个目标来弥补这一差距： 在目标1中，我们将开发计算工具，以允许对MRI数据进行多方面的分析，包括体积和 白质/软膜表面，将在公共空间进行，以便更全面地了解早期 大脑发育我们的工具将明确考虑MR图像外观的快速变化，这些变化是典型的， 生命的第一年。与设计用于图像体积或皮质表面的常规方法不同， 导致不一致和对细微变化的敏感性丧失，我们的工具将允许联合体积-表面 在一致的纵向空间中进行分析。通过从两者提取信息来提高配准精度 实体对于检测发育中的大脑的细微变化至关重要，大脑明显较小， 大脑皮层 在目标2中，我们将为早期发育的儿童生成纵向、多模态和全脑分区图。 个脑袋将大脑细分为连贯的区域是宏观绘制spa的重要步骤。 基本上异质的变化和空间和拓扑组织的检查。我们的方法将 允许表征分块在不同时间的演变，同时保持时间 包裹的一致性和主体间的对应性。 在目标3中，我们将开发技术，允许预测缺失的MRI数据，以增加 不完整的数据，以提高统计能力。缺失数据是纵向研究中常见且不可避免的问题 由于受试者脱落或扫描失败而导致的研究，尤其是涉及婴儿的研究。为了解决这个问题， 我们将开发深度学习技术，用于丢失成像数据的纵向预测。 该项目的成功完成将为神经科学界提供更多的计算工具， 使用MRI精确绘制人类大脑规范早期发育的图表。作为该项目的一部分，我们将 提供第一套时间密集的表面体积地图集，以捕捉关键的发育特征 因此对于量化可能偏离正常大脑发育的情况至关重要。

Longitudinal Mapping of Human Brain Development in the First Years of Life Abstract This proposal requests continued funding support for research at the University of North Carolina at Chapel Hill to develop computational tools for quantifying longitudinal structural changes in the human brain. The previous project period has been extremely successful in advancing robust tools for longitudinal brain analysis of the aging brain. In this renewal, we seek to further advance robust computational tools for comprehensive longitudinal characterization of changes in the early developing brain. This is in line with our long-term goal of creating computational tools for longitudinal charting of brain evolution across the entire human lifespan. The tools to be developed in this project will allow unified and concurrent analysis of longitudinal volumetric data and cortical surfaces, facilitating the mapping of dynamic and spatially heterogeneous structural changes during a critical period of brain development. The tools developed in this project will be tailored to studying the human brain in the first few years of life, which undergoes dynamic development in both structure and function. We will utilize the MRI data made available via the Baby Connectome Project (BCP), involving 500 pediatric subjects scanned from birth to five years of age. The outcome of BCP will inform neuroscientists what normal healthy growth looks like and facilitate discovery of the earliest manifestations of brain disorders. To fully benefit from this unique dataset, dedicated computational tools are needed for accurate processing and analysis of baby MR images, which typically exhibit dynamic heteroge- neous changes across time. However, most computational tools developed to date have been mostly focused on adult subjects and are unreliable when applied to baby MRI. We propose to address this gap with three aims: In Aim 1, we will develop computational tools to allow multifaceted analysis of MRI data, including volumes and white-matter/pial surfaces, to be carried out in common spaces for a more holistic understanding of the early developing brain. Our tools will explicitly consider the rapid changes in MR image appearances that are typical in the first year of life. Unlike conventional methods that are designed for either image volumes or cortical surfaces, resulting in inconsistencies and loss of sensitivity to subtle changes, our tools will allow joint volume-surface analysis in consistent longitudinal spaces. Improving registration accuracy by drawing information from both entities is critical for detecting subtle changes in the developing brain, which is significantly smaller with a thinner cerebral cortex. In Aim 2, we will generate longitudinal, multimodal, and whole-brain parcellation maps for the early developing brain. Subdivision of the brain into coherent regions is an essential step in the macroscopic mapping of spa- tially heterogeneous changes and in the examination of spatial and topological organization. Our approach will allow the characterization of the evolution of parcellation across time and at the same time maintain temporal consistency and inter-subject correspondences of the parcels. In Aim 3, we will develop techniques that will allow prediction of missing MRI data to increase the usability of incomplete data for improving statistical power. Missing data is a common and inevitable problem in longitudinal studies due to subject dropouts or failed scans, especially in studies involving infants. To address this problem, we will develop deep learning techniques for longitudinal prediction of missing imaging data. Successful completion of this project will empower the neuroscience community with computational tools for more precise charting of the normative early development of the human brain using MRI. As part of this project, we will deliver the first set of temporally-dense surface-volumetric atlases that will capture key developmental traits and are therefore critical for quantification of possible deviation from normal brain development.