Statistical methods to improve reproducibility and reduce technical variability in heterogeneous multimodal neuroimaging studies of Alzheimer’s Disease

项目摘要/摘要：阿尔茨海默病(AD)是一种具有史诗般的负担的公共卫生危机 考虑到仅2018年美国社会的估计成本就达2770亿美元，这一比例。脑成像 结合新的形态计量分析方法，从根本上改变了我们对AD的理解 进步。然而，进展已经放缓，因为AD的大脑表现出实质性的萎缩，白色 物质病理学和大变形，这使得最常用的软件很难 进行AD患者纵向研究严重依赖的组织分割。我们 建议开发新的、可推广和可重复性的统计神经成像前处理方法 专门为高度异质的AD MRI/PET图像人群量身定做，并随后评估这些 方法相对于标准方法。具体地说，我们将重点介绍组织类分割，这通常是 直接用于统计分析或作为空间或多式联运登记的中介步骤，因为我们 评估标准软件在异质性AD和老年人中的组织分类分割性能 控制研究人群。 这个项目的主要目标是产生改进的、可重现的和开放源码的统计方法 AD患者和老年对照组的组织分类分割。为了实现这一目标，我们提出了三个主要建议 假设：1)提出了一种新的组织类分割方法 对健康对照组、阿尔茨海默病患者和健康老年对照组的纵向研究；2)将方法扩展到 考虑不同的研究和实验条件(例如，核磁共振扫描仪)并评估其重复性 用于年轻健康对照和AD受试者的结构MRI和PET，以及3)开发在线、免费访问、 可重复使用的软件工具，用于对已公布的分析管道进行评估、验证和再现性。 这项研究的完成将为神经影像临床研究的分析提供强有力的工具 来自阿尔茨海默病受试者。这项工作将有助于验证，重复性和实验设计通过改进 现有的分析技术可以准确地量化生物标记物和治疗对AD脑病理的影响。

Project Summary/Abstract: Alzheimer's disease (AD) is a public health crisis with a burden of epic proportion on the American society given its estimated cost of $277 billion in 2018 alone. Brain imaging combined with new morphometric analytic methods has fundamentally changed our understanding of AD progression. However, progress has been slowed because the AD brain exhibits substantial atrophy, white matter pathology, and large deformations, which make it difficult for the most commonly used software package to carry out the tissue segmentation on which longitudinal studies of AD patients depend heavily. We propose to develop novel, generalizable and reproducible statistical neuroimaging pre-processing methods tailored specifically for highly heterogeneous AD MRI/PET image populations and to subsequently assess these methods relative to standard approaches. Specifically, we will focus on tissue class segmentation, which is often used directly for statistical analyses or as an intermediary step for spatial or multimodal registration, as we evaluate the performance of standard software for tissue class segmentation in a heterogeneous AD and elderly control study population. The primary goal of this project is to produce improved, reproducible, and open source statistical methods for tissue class segmentation for AD patients and elderly controls. To achieve this goal we propose three main hypotheses: 1) develop new tissue class segmentation methods for heterogeneous cross-sectional and longitudinal studies of healthy controls, AD subjects and healthy elderly controls; 2) extend the methods to account for different studies and experimental conditions (e.g., MRI scanner) and evaluate their reproducibility for structural MRI and PET in young healthy controls and AD subjects and 3) develop online, freely accessible, reproducible software tools for the assessment, validation, and reproducibility of published analytic pipelines. The completion of this research will provide powerful tools for the analysis of neuroimaging clinical studies from subjects with AD. This work will aid in validation, reproducibility and experimental design by improving existing analysis techniques to accurately quantify biomarkers and treatment impact on brain pathology in AD.