Harmonizing multi-site diffusion MRI acquisitions for neuroscientific analysis across ages and brain disorders

摘要（ ! 扩散MRI（dMRI）是唯一一种可以绘制活体人脑连接的非侵入性方法， 对理解精神疾病至关重要几项大型研究，如人类连接组计划（HCP） 和青少年大脑认知发展（ABCD）已经收集或准备收集扩散MRI 超过30，000名受试者的数据。然而，一个重要的挑战是，这些数据集收集自不同的 由于扫描仪间（研究中心间）差异较大，无法合并扫描仪进行联合分析， 用于数据重建的供应商特定软件的差异、头部线圈的灵敏度等。这些扫描仪 在精神疾病组之间观察到的差异往往大于效应量。第二 大规模数据分析的挑战是缺乏一个一致的基于本体的定义和自动化的 提取整个生命周期（包括新生儿和儿童）的白色物质连接。第三个挑战是 组合的dMRI数据集的绝对大小（几TB），限制了研究人员测试的能力 假设，因为这需要专业知识和复杂的计算资源来处理、存储和 可视化如此大量的数据。在这项赠款中，我们建议解决这些挑战，使大- dMRI数据的规模数据密集型分析。具体而言，在目标1中，我们建议开发新的数学 从多个站点获取的数据中去除扫描仪特定差异的算法。我们将协调10，000 来自ABCD研究的受试者来自21个不同的研究中心，另外10，000名受试者来自HCP倡议 跨越整个生命周期和许多疾病适应症，10，000名健康大脑受试者 网络所有协调数据集（30，000名受试者）将使用NIMH数据与社区共享 存档（NDA）。在目标2中，我们将开发一个正式的基于本体的系统，用于定义189个白色物质束 使用从人类和猴子关于大脑连接的文献中已知的神经解剖学标志。我们的主要 重点将是开发新的算法，自动和一致的聚类和提取这些纤维 包括新生儿在内的整个人类生命周期。为了能够广泛使用， 要求计算资源和技术知识，在目标3中，我们将开发一个基于Web的系统 用于协调数据和分册的实时3D查看和查询（与NIMH数据存档集成 基础设施），用于从不同的受试者的整个队列中用户定义的受试者选择。 诊断类别。总的来说，这一框架的潜在影响是巨大的，因为它将首次， 允许对dMRI数据进行大规模的数据密集型分析，以研究神经发育和精神障碍 跨越诊断界限。 !

Abstract( ! Diffusion MRI (dMRI) is the only non-invasive method that can map the living human brain’s connections and is critical for understanding mental disorders. Several large studies such as the Human Connectome Project (HCP) and the Adolescent Brain Cognitive Development (ABCD) have collected or are poised to collect diffusion MRI data from over 30,000 subjects. However, an important challenge is that these datasets collected from different scanners cannot be pooled for joint analysis due to large inter-scanner (inter-site) differences, caused by differences in vendor specific software for data reconstruction, the sensitivity of head coils etc. These scanner differences are often larger than the effect sizes observed between groups in psychiatric disorders. A second challenge for large-scale data analysis is the lack of a single consistent ontology-based definition and automated extraction of white matter connections across the lifespan (including neonates and children). A third challenge is the sheer size of the combined dMRI datasets (several terabytes), limiting the ability of researchers to test hypotheses as this requires expertise and complex computational resources for processing, storing, and visualizing such large volumes of data. In this grant, we propose to address these challenges to enable large- scale data-intensive analysis of dMRI data. Specifically, in Aim 1, we propose to develop novel mathematical algorithms to remove scanner-specific differences from data acquired at multiple sites. We will harmonize 10,000 subjects from the ABCD study acquired at 21 different sites, another 10,000 subjects from the HCP initiative spanning the entire lifespan and numerous disease indications and 10,000 subjects from the Healthy Brain Network. All the harmonized datasets (30,000 subjects), will be shared with the community using the NIMH data archive (NDA). In Aim 2, we will develop a formal ontology-based system for defining 189 white matter fascicles using neuroanatomical landmarks known from human and monkey literature on brain connectivity. Our main focus will be to develop novel algorithms for automated and consistent clustering and extraction of these fiber bundles spanning the entire human lifespan including neonates. To enable widespread use without the need for demanding computational resources and technical knowledge, in Aim 3, we will develop a web-based system for real-time 3D viewing and querying of the harmonized data and fascicles (integrating with NIMH data archive infrastructure) for a user-defined selection of subjects from the entire cohort of subjects across different diagnostic categories. Overall, the potential impact of this framework is significant, as it will, for the first time, allow a large-scale data-intensive analysis of dMRI data to study neurodevelopment as well as mental disorders cutting across diagnostic boundaries. !