FiberNET: Deep learning to evaluate brain tract integrity worldwide and in AD

项目资助-项目概要/摘要 阿尔茨海默病（Alzheimer's disease，AD）是一种严重威胁人类健康的疾病。65岁以上每满5岁， AD的患病率翻了一番，2018年美国的成本估计为2770亿美元，比2018年增加了200亿美元。 上年在这里，我们提出了一个协调的全球研究大脑老化和AD，使用新的方法 来评估大脑神经通路的白色微观结构--这是一个关键的大脑指标， 从分子AD病理学到临床衰退的途径。借助一种名为FiberNET的新型深度学习工具， 我们提取并分析了大脑的白色物质纤维束，这些纤维束是从扩散MRI（dMRI）扫描中获得的， 世界，并回答3个关键问题：大脑的神经束如何在世界范围内老化？道老化如何取决于 老年痴呆症的遗传风险和大脑淀粉样蛋白含量道指标能否更好地预测临床下降，当 与标准的、公认的AD生物标志物相结合？该提案联合了AD，神经成像，机器 学习和大规模基因组学，将新的衰老指标（道微结构）与保护性和不良反应联系起来， 因素新的数学包括创新，在挑选交叉纤维和组织性质，从多个， 壳扩散MRI和卷积神经网络来学习世界范围内神经通路的老化模式。我们 目标是（1）使用FiberNET，我们的深度学习方法，从全球范围内的大脑dMRI扫描中提取神经束， 在20，000人的整个生命周期中创建正常道老化的标准图表;（2）询问道老化是如何发生的。 轨迹取决于AD保护基因型APOE 2、风险基因型APOE 4和脑淀粉样蛋白负荷 用淀粉样蛋白敏感的PET测量。拟议的研究将创建白色物质束的标准化图表 完整性作为规范性白色物质老化的路标。我们建立在我们的谜- Lifespan工作-分析了来自91个队列的10，144人的大脑MRI数据-以创建寿命图表 对于来自dMRI的大脑主要神经束，产生用于AD组比较的基本规范性信息 国际吧这种寿命方法将有助于发现加速衰老的个人因素， 人口标准（例如，APOE基因型和淀粉样蛋白负荷）。为了确保事态发展的影响，我们 创建了一个beta测试团队来帮助测试和改进这些方法，这些方法紧密地集成到我们的ENIGMA中。 该联盟致力于跨队列数据协调。这种全球性的老龄化和AD方法将 提供了一个新的力量来源，以“打破僵局”，发现影响大脑的因素，因为我们的年龄。

PARENT GRANT - PROJECT SUMMARY/ABSTRACT Alzheimer’s disease (AD) threatens to devastate society worldwide. For every 5 years of age over age 65, the prevalence of AD doubles, costing an estimated $277 billion in the U.S. in 2018, a $20 billion increase from the previous year. Here we propose a coordinated global study of brain aging and AD that uses novel approaches to assess the white matter microstructure of the brain’s neural pathways - a crucial brain metric that breaks down on the pathway from molecular AD pathology to clinical decline. With a novel deep learning tool, called FiberNET, we extract and analyze the brain’s white matter fiber bundles obtained from diffusion MRI (dMRI) scans across the world, and answer 3 key questions: how do the brain’s tracts age worldwide? How does tract aging depend on Alzheimer’s genetic risk and brain amyloid load? Can tract metrics predict clinical decline better, when combined with standard, accepted biomarkers of AD? The proposal unites experts in AD, neuroimaging, machine learning, and large-scale genomics, to relate new aging metrics (tract microstructure) to protective and adverse factors. Novel mathematics include innovations in picking up crossing fibers and tissue properties from multi- shell diffusion MRI, and convolutional neural nets to learn patterns of aging in neural pathways worldwide. We aim to (1) use FiberNET, our deep learning method, to extract tracts from brain dMRI scans worldwide, and create normative charts for normal tract aging in 20,000 people across the lifespan; (2) ask how the tract aging trajectory depends on the AD protective genotype APOE2, risk genotype APOE4, and brain amyloid load measured with amyloid-sensitive PET. The proposed study will create standardized charts of white matter tract integrity across the lifespan to serve as a guidepost for normative white matter aging. We build on our ENIGMA- Lifespan work - which analyzed brain MRI data from 10,144 people from 91 cohorts - to create lifespan charts for the brain’s major tracts from dMRI, yielding fundamental normative information for comparisons of AD groups worldwide. This lifespan approach will aid the discovery of personal factors that accelerate aging relative to population norms (e.g., APOE genotype, and amyloid load). To ensure the impact of the developments, we created a team of beta-testers to help test and refine the methods, that is tightly integrated into our ENIGMA consortium, which is dedicated to cross-cohort data harmonization. This global approach to aging and AD will offer a new source of power to “break the logjam” in discovering factors that affect the brain as we age.