TR&D2: Diffusion MRI and Connectomics

TR

• 批准号: 10427164
• 负责人: PAUL M THOMPSON
• 金额: $ 27.89万
• 依托单位: UNIVERSITY OF SOUTHERN CALIFORNIA
• 依托单位国家: 美国
• 财政年份: 1998
• 资助国家: 美国
• 起止时间: 1998-09-30 至 2024-02-29
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10427164
• 关键词: Address; Adolescent; Affect; Algorithms; Anatomy; Anisotropy; Area; Atlases; Brain; Brain Diseases; Brain region; Child; Clinical Research; Collaborations; Data; Data Set; Dementia; Detection; Development; Diffusion; Diffusion Magnetic Resonance Imaging; Disease; Early Onset Alzheimer Disease; Elderly; Fiber; Frontotemporal Dementia; Goals; HIV; Head; Image; Intervention; Label; Laboratory of Neuro Imaging Resource; Learning; Longevity; MRI Scans; Machine Learning; Manuals; Maps; Mathematics; Measures; Medicine; Mental Depression; Mental disorders; Methods; Minnesota; Modeling; Nerve Degeneration; Neurosciences; Paper; Pathway Analysis; Pathway interactions; Patients; Pattern; Persons; Population; Post-Traumatic Stress Disorders; Property; Publishing; Reproducibility; Research Personnel; Risk; Running; Scanning; Site; Skeleton; Students; Testing; Thailand; Work; base; biophysical model; blast exposure; brain tissue; connectome; deep learning; density; detection sensitivity; high dimensionality; improved; innovation; learning strategy; mild traumatic brain injury; neuropsychiatric disorder; novel; novel strategies; spectrograph; statistics; tool; tractography; ultra high resolution; white matter

PROJECT SUMMARY- TR&D2: DIFFUSION MRI AND CONNECTOMICS In our new TR&D2 project, Diffusion Imaging & Connectomics, we extend our decades of work in the mathematics of diffusion MRI to create new and useful tools. Our work draws on our last 4 years of productive collaborations, which led to over 100 papers. We focus on 3 tasks: (1) modeling white matter microstructure, (2) automatically extracting maps of fiber bundles, and (3) modeling brain connectivity in more powerful and adaptive ways. Each method seeks to overcome a major problem that affects the validity of diffusion MRI today. In Aim 1, we develop mathematical metrics of white matter microstructure that can better use the available information in low-quality dMRI, but drastically boost the analytical power of higher-quality dMRI. With our diverse range of Collaborative Projects - including one that collects ultra-high resolution diffusion spectrum imaging at high-field (7T DSI - we test extensions of our tensor distribution function model of dMRI, which outperforms the standard tensor model and the most popular metric, DTI-FA. We extend the TDF to bi-exponential and multi- compartmental models, using TV-L1 fusion to merge all metrics across the image to better detect disease. In Aim 2, we extend our tools for automatic tract labeling. Our new approach, FiberNET, uses deep learning to avoid the manual intervention required by our initial tool, autoMATE. We will compare it head to head for accuracy, power, and utility on all our Collaborative Project datasets. These include diffusion MRI data from HIV+ children, from elderly people with various types of dementia, and adolescents at risk for psychiatric disorders. In Aim 3, we use two approaches - L1 fusion (dFUSE) and connectivity based on continuous functions (ConCon) - to overcome 2 very serious problems with the analysis of brain networks - the problem of large number of false positive fibers in tractography, and the arbitrariness of picking a cortical parcellation. These problems affect all downstream network metrics, to the point where the networks are very hard to compare across studies, with current methods. ConCon models the connectivity directly as a density, making it amenable to registration and statistics without defining specific borders in the cortex; dFUSE combines the reliable fibers across many tractography methods, boosting the contribution of methods that are more reliable on a given dataset. All our tools will be widely disseminated. With our broad collaborative network, we hope to advance the neuroscience of brain connectivity by using novel mathematics, tested head to head against existing methods.

项目摘要 - TR&D2：扩散 MRI 和连接组学 在我们新的 TR&D2 项目“扩散成像和连接组学”中，我们延续了数十年的工作成果 扩散 MRI 的数学创造了新的有用的工具。我们的工作借鉴了过去 4 年的富有成效的成果 合作，发表论文 100 多篇。我们专注于 3 项任务：(1) 建模白质微观结构，(2) 自动提取纤维束图，以及（3）以更强大和适应性更强的方式对大脑连接进行建模 方式。每种方法都试图克服影响当今扩散 MRI 有效性的主要问题。在目标 1 中， 我们开发了白质微观结构的数学度量，可以更好地利用现有信息 低质量 dMRI，但大大提高了高质量 dMRI 的分析能力。凭借我们多样化的 合作项目 - 包括在高场收集超高分辨率扩散光谱成像的项目 （7T DSI - 我们测试了 dMRI 张量分布函数模型的扩展，其性能优于标准 张量模型和最流行的度量 DTI-FA。我们将 TDF 扩展到双指数和多指数 隔室模型，使用 TV-L1 融合来合并图像中的所有指标，以更好地检测疾病。在 目标 2，我们扩展了自动管道标记工具。我们的新方法 FiberNET 使用深度学习来 避免我们最初的工具 autoMATE 所需的手动干预。我们将对其进行头对头比较以确保准确性， 我们所有协作项目数据集的力量和实用性。其中包括来自 HIV+ 儿童的扩散 MRI 数据， 来自患有各种类型痴呆症的老年人和有精神疾病风险的青少年。在目标 3 中， 我们使用两种方法 - L1 融合 (dFUSE) 和基于连续函数的连接 (ConCon) - 克服脑网络分析的2个非常严重的问题——大量错误的问题 纤维束成像中的阳性纤维，以及挑选皮质分区的任意性。这些问题影响着所有人 下游网络指标，以至于网络很难在研究之间进行比较， 目前的方法。 ConCon 将连通性直接建模为密度，使其适合配准和 没有定义皮层特定边界的统计数据； dFUSE 结合了多种可靠的光纤 纤维束成像方法，提高了在给定数据集上更可靠的方法的贡献。我们所有的 工具将得到广泛传播。凭借我们广泛的合作网络，我们希望推动神经科学的发展 通过使用新颖的数学方法来研究大脑连接性，并与现有方法进行正面测试。