Leveraging artificial intelligence to develop novel tools for studying infant brain development

项目总结。人类生命的头24个月是充满活力的，以快速增长为特征，并且 越来越多的人认识到，这对建立持续一生的认知能力和行为至关重要。然而， 对大脑结构和功能发育在这一敏感期的轨迹知之甚少。 通常是发育中的婴儿，而对于这些轨迹中的偏差与出现的关系更是知之甚少 认知和行为或预测以后的发展结果。这在一定程度上是由于目前的技术 磁共振成像(MRI)定量测定婴儿脑结构和功能的局限性 发展神经科学研究的重要、非侵入性方法。目前，有足够的 方法分析婴儿出生后24个月的MRI扫描，特别是脑部分割- 第一步，也是关键的一步，用于几乎所有MRI模式的定量分析。如果没有准确和自动化的 分割，婴儿MRI分析容易出现系统误差，而且是劳动密集型的，限制了严谨和 婴儿核磁共振研究的重复性。这一限制限制并推迟了大规模婴儿核磁共振的应用 在可预见的未来的数据集。解决这些研究差距将极大地推动以下努力 早期识别发育迟缓和/或障碍。我建议发展基于人工智能的婴儿神经成像 通过两个大规模数据集研究人类早期大脑发育的分析工具：美国国立卫生研究院资助的婴儿 Connectome计划和来自环境对儿童健康影响的集中式MRI数据存储库 结果。在我的初步研究中，我已经证明了这部剧与来自 两个不同的来源，与其他常用的分割方法相比，性能更优越。我的 第一个目标是开发一种使用3D卷积神经网络的自动化和可推广的脑分割流水线 网络--一种人工智能方法。这种分割工具可以适应和处理婴儿的脑部扫描 在生命的头两年里，每个月都有。最终的基于人工智能的管道将在内部进行严格验证，以及 经过外部测试。我们将把管道作为用户友好的、基于Web的界面发布，供研究人员在 科学界。在目标2中，我将描绘区域脑形态计量学的生长轨迹，主要 功能网络，并测量它们与前24个月神经心理功能的关系 通过来自BCP的数据了解生命。在目标3中，我将利用两种不同的方法(人工智能和LPCA)来预测 评估3岁以下儿童的发育结果。在第一年的纵向多模式MRI扫描中 BCP。该奖项的跨学科培训阶段，在乔纳森·波斯纳博士的实验室进行， 哥伦比亚大学，包括一项全面的技术和专业技能获取计划， 将使我能够过渡到研究独立。这一项目的圆满完成将产生可靠的 研究早期大脑发育的工具和新颖的数据驱动方法，填补早期关键知识空白 发展，并推动努力及早发现发育迟缓和障碍。

PROJECT SUMMARY. The first 24-months of human life are dynamic, characterized by rapid growth, and increasingly recognized as crucial for establishing cognitive abilities and behaviors that last a lifetime. However, little is known about trajectories of structural and functional brain development during this sensitive period in typically developing infants, and even less is known about how deviations in these trajectories relate to emerging cognition and behavior or predict later developmental outcomes. This is partially due to current technical limitations on quantification of brain structure and function in infants via magnetic resonance imaging (MRI) – an important, non-invasive approach to the study of developmental neuroscience. Currently there are insufficient methods to analyze infant MRI scans across the first 24 months of life, especially for brain segmentation – the first and critical step for virtually all quantitative analyses across MRI modalities. Without accurate and automated segmentation, infant MRI analysis is prone to systematic errors and is labor-intensive, limiting the rigor and reproducibility of infant MRI research. This limitation curtails and delays the utility of large-scale infant MRI datasets in the foreseeable future. Addressing these research gaps would significantly advance efforts toward early identification of developmental delays and/or disorders. I propose developing AI-based infant neuroimaging analysis tools for studying the early human brain development via two large-scale datasets: the NIH funded Baby Connectome Project and a centralized MRI data repository from Environmental Influence on Child Health Outcomes. In my pilot studies, I have shown the show good-to-excellent agreement with ground-truth labels from two different sources, and superior performance compared to other commonly used segmentation methods. My first aim is to develop an automated and generalizable brain segmentation pipeline with 3D convolutional neural networks – an AI approach. This segmentation tool can accommodate and process infant brain scans spanning each month over the first 2 years of life. The final AI-based pipeline will be rigorously validated internally, and tested externally. We will release the pipeline as a user-friendly, web-based interface for researchers to use in scientific community. In Aim 2, I will delineate the growth trajectories of regional brain morphometrics, major functional networks, and measure their relationships to neuropsychological functions during the first 24months of life via data from BCP. In Aim 3, I will leverage two different approaches (AI and LPCA) to predict the developmental outcomes assessed up to 3 years old. with the first-year longitudinal multimodal MRI scans from BCP. The interdisciplinary training phase of the award, conducted in the laboratory of Dr. Jonathan Posner at Columbia University, includes a comprehensive plan for the acquisition of technical and professional skills that will enable my transition to research independence. The successful completion of this project will yield reliable tools and novel data-driven methods for studying early brain developmental, fill critical knowledge gaps of early development, and advance efforts toward early identification of developmental delays and disorders.