Harmonization for multisite Connectomics: parsing heterogeneity and creating markers in ASD

多站点连接组学的协调：解析 ASD 中的异质性并创建标记

基本信息

批准号：
9927671
负责人：
Ragini Verma
金额：
$ 76.03万
依托单位：
UNIVERSITY OF PENNSYLVANIA
依托单位国家：
美国
项目类别：
财政年份：
2019
资助国家：
美国
起止时间：
2019-05-08 至 2024-01-31
项目状态：
已结题

来源：
https://reporter.nih.gov/project-details/9927671
关键词：
Address Affect Age Anisotropy Artificial Intelligence Attention deficit hyperactivity disorder Behavioral Genetics Biological Biological Markers Biology Brain Brain region Clinical Cognitive Communication Communities Custom Data Data Set Diffuse Diffusion Magnetic Resonance Imaging Dimensions Disease Effectiveness Future Goals Heterogeneity Image Individual Investigation Learning Link Magnetic Resonance Imaging Measurement Measures Mental Health Methods Modality Mood Disorders Morphologic artifacts Motion Patients Phenotype Protocols documentation Psychotic Disorders Quality Control Research Sample Size Sampling Scanning Site Source Structure Subgroup Symptoms System Tissues Training autism spectrum disorder base data harmonization data integration deep learning design diagnostic biomarker indexing large datasets neuroimaging precision medicine preservation repetitive behavior sex social success tool white matter

项目摘要

Diffusion MRI (dMRI) provides a superior characterization of white matter and connectivity compared to other MRI modalities, and is routinely included in studies of disorders with atypical brain connectivity like autism spectrum disorder (ASD). The field could benefit tremendously from combining studies, to have comprehensive representation of the underlying heterogeneity in connectivity-based disorders. This is rendered challenging by dMRI being very sensitive to acquisition parameters, needing sophisticated statistical harmonization tools due to the complicated effect of scanner related changes. This also calls for a robust automated quality control (QC) protocol prior to data harmonization. Thus, in this proposal, we will develop tools to facilitate integration of dMRI data across studies. In Aim 1, we will develop and validate a deep learning based tool for automating QC for dMRI data that will identify different data artifacts (caused by multiple sources like scanner, coil, scan parameters, motion etc), and the appropriate action that needs to be taken (like motion and eddy correction). In Aim 2, we will develop a suite of tools for harmonizing dMRI measures to remove acquisition differences. The effectiveness of our proposed tools will be demonstrated by harmonizing ~1500 datasets (ages 6-32 years) from 11 ASD studies. These large harmonized datasets create the need for a subject-wise characterization of the sample and for diagnostic markers that harness the imaging heterogeneity of the larger harmonized sample. To address this new need, we will develop additional connectomic analysis tools, that will be adapted to ASD to create the CHARM (Connectomic Heterogeneity in Autism Research through Multi-site dMRI harmonization) suite comprising of a generalizable biomarker of ASD, as well as a dimensional connectomic coordinate system. In Aim 3, we will characterize each subject using a connectivity phenotype, cluster the integrated ASD sample based on this connectivity-phenotype, define a classifier for each cluster; and create a connectivity-based ensemble biomarker of ASD, called the CHARM-marker, combining these cluster-specific classifier decisions. Finally, in Aim 4, we will create a subject-wise characterization of ASD by designing a multi-dimensional connectomic coordinate system using metric learning, to quantify the dissimilarity of each subject from the harmonized healthy controls. We will elucidate the link of these CHARM-coordinates to ASD constructs, by correlating core ASD symptoms with the CHARM coordinates in the harmonized/combined sample. This will enable the ASD community to associate informative connectomic dimensions with each subject, facilitating subject-wise longitudinal assessment, paving the way for precision medicine. Such a group- based and subject-wise characterization of ASD could not have been possible without data integration. Additionally, the neuroimaging community will have new dMRI harmonization and connectomic analysis tools enabling the integration of studies for a more comprehensive connectomic investigation of existing data. It will pave the way for such studies in other connectivity-related disorders that affect mental health.

与其他技术相比，扩散 MRI (dMRI) 可以更好地表征白质和连接性 MRI 模式，通常包含在自闭症等非典型大脑连接疾病的研究中谱系障碍（ASD）。该领域可以从结合研究中获益匪浅，获得全面的研究成果基于连通性的疾病中潜在异质性的表示。这变得具有挑战性 dMRI 对采集参数非常敏感，需要复杂的统计协调工具，因为扫描仪相关变化的复杂影响。这也需要强大的自动化质量控制数据协调之前的（QC）协议。因此，在本提案中，我们将开发工具来促进集成跨研究的 dMRI 数据。在目标 1 中，我们将开发并验证基于深度学习的自动化 QC 工具用于识别不同数据伪影的 dMRI 数据（由扫描仪、线圈、扫描仪等多个源引起）参数、运动等），以及需要采取的适当措施（如运动和涡流校正）。在目标 2，我们将开发一套工具来协调 dMRI 测量，以消除采集差异。这我们提出的工具的有效性将通过协调约 1500 个数据集（年龄 6-32 岁）来证明来自 11 项 ASD 研究。这些大型统一数据集需要按主题描述样本和诊断标记物利用更大的统一的成像异质性样本。为了满足这一新需求，我们将开发额外的连接组分析工具，这些工具将适应与 ASD 合作创建 CHARM（通过多站点 dMRI 进行自闭症研究中的连接组异质性）协调）套件包含 ASD 的通用生物标志物以及维度连接组坐标系。在目标 3 中，我们将使用连通性表型来表征每个受试者，对基于该连通性表型的集成 ASD 样本，为每个簇定义一个分类器；并创建一个基于连通性的 ASD 整体生物标记，称为 CHARM 标记，结合了这些簇特定的标记分类器决策。最后，在目标 4 中，我们将通过设计一个使用度量学习的多维连接体坐标系，以量化每个连接体的差异受统一健康控制的影响。我们将阐明这些 CHARM 坐标与 ASD 的联系通过将核心 ASD 症状与协调/组合中的 CHARM 坐标相关联来构建样本。这将使 ASD 社区能够将信息丰富的连接体维度与每个主题，促进主题纵向评估，为精准医学铺平道路。这样的一个群体—— 如果没有数据集成，就不可能对 ASD 进行基于主题的表征。此外，神经影像学界将拥有新的 dMRI 协调和连接组分析工具能够整合研究，以便对现有数据进行更全面的连接组学调查。它将为影响心理健康的其他连接相关疾病的此类研究铺平道路。