Fluency from Flesh to Filament: Collation, Representation, and Analysis of Multi-Scale Neuroimaging data to Characterize and Diagnose Alzheimer's Disease

项目总结/摘要 诊断、理解和治疗阿尔茨海默病（AD）的主要障碍是其 通过tau蛋白和β-淀粉样蛋白（A β）病理学模式表征，仅通过传统的 组织切片和染色方法。为解决这一问题，2018年框架之后的最新努力 由国家老龄化研究所（NIA）和阿尔茨海默氏症协会（AA）提出的研究重点是 鉴定可替代用于表征AD的体内生物标志物，特别是沿着连续体。 从MRI收集的测量，如皮质厚度，构成了一类这样的生物标志物。而 它们已被证明与AD的临床分期相关，MRI生物标志物尚未被证明与AD的临床分期相关。 因为它们不能与AD的tau/A β的特征模式联系起来， 计算工具和建模框架。本项目的目标是通过 开发和实施多模式、多尺度图像配准和分析平台， 用于将显微病理学数据与宏观MRI测量进行整合并在统计学上关联， 皮质厚度约翰霍普金斯大脑资源和AD研究中心已经准备了2D数字 图10显示了来自以下患者的针对tau（PHF-1）染色的组织学图像和相应的内侧颞叶（MTL）组织的3D MRI： 对照组和中晚期AD患者的大脑。用免疫组织化学方法检测单个tau缠结。 基于卷积神经网络（UNET）的方法，在手动注释的组织学特征的子集上训练 样品将MRI手动分割成MTL区域，并从以下位置测量皮质厚度： 从这些区域中的每一个的生成的表面表示。该项目的总体目标将是 通过两个主要目标实现。首先，tau缠结和皮质厚度测量将被共定位在 Mai-Paxinos地图集的坐标空间，通过开发配准算法，使用1） 多目标模型，用于解释组织学图像和MRI中可能的失真，2）“散射” 转换”，以捕捉组织学图像中的纹理特征，帮助预测灰色与 白色物质，3）非刚性转换的区域表面表示的Mai-Paxinos地图集。 第二，将使用层次结构计算tau缠结和皮质厚度之间的统计相关性。 捕获数据值和相对组织面积以说明尺度差异的“可变”测量 （微观与宏观）和采样频率（不规则与规则）这两个数据集。应用 这些方法的控制和AD脑样品将表征皮质厚度的相关性 测量沿着AD的临床连续体和在3D空间中的物理上的tau缠结密度， MTL的区域，以及沿着大脑的沿着特定轴。这些相关性将表征 AD的皮质厚度测量，通过开源平台共享这些方法将使 这种表征在将来用于其他MRI生物标志物。

Project Summary/Abstract A major obstacle in diagnosing, understanding, and treating Alzheimer’s Disease (AD) has been its characterization by patterns of tau and beta-amyloid (Aß) pathology, only adequately seen through traditional methods of histological sectioning and staining. To address this, recent efforts following the 2018 framework put forth by the National Institute of Aging (NIA) and the Alzheimer’s Association (AA) have focused on identifying in vivo biomarkers that can be used instead to characterize AD and specifically along a continuum. Measures gleaned from MRI, such as cortical thickness, constitute one category of such biomarkers. While they have been shown to correlate with clinical stage of AD, MRI biomarkers have not been shown to be specific for AD as they have not been able to be linked to AD’s signature patterns of tau/Aß with current computational tools and modeling frameworks. The goal of this project is to address this deficiency with the development and implementation of a multi-modal, multi-scale image registration and analysis platform that will be used to integrate and statistically correlate microscopic pathology data with macroscopic MRI measures of cortical thickness. The Johns Hopkins Brain Resource and AD Research Centers have prepared 2D digital histology images stained for tau (PHF-1) and corresponding 3D MRI of medial temporal lobe (MTL) tissue from control brains and those with intermediate and advanced AD. Individual tau tangles were detected with a convolutional neural network (UNET) based approach trained on a subset of manually annotated histological samples. MRI was manually segmented into regions of the MTL, and cortical thickness will be measured from from generated surface representations of each of these regions. The project’s overall goal will be accomplished through two main aims. First, tau tangle and cortical thickness measures will be co-localized in the coordinate space of the Mai-Paxinos Atlas through the development of a registration algorithm that uses 1) a multi-target model to account for possible distortion in both histology images and MRI, 2) a “Scattering Transform” to capture textural features in histology images that help predict delineations between grey vs. white matter, 3) non-rigid transformation of regional surface representations to those of the Mai-Paxinos Atlas. Second, statistical correlations will be computed between tau tangles and cortical thickness using a hierarchy of “varifold” measures that capture both data values and relative tissue area to account for differences in scale (microscopic vs. macroscopic) and sampling frequency (irregular vs. regular) of these two datasets. Application of these methods to both control and AD brain samples will characterize the correlation of cortical thickness measures to tau tangle density along the clinical continuum of AD and physically in 3D space, within specific regions of the MTL, and along particular axes of the brain. These correlations will characterize the specificity of cortical thickness measures for AD, and the sharing of these methods via an open-source platform will enable this characterization for other MRI biomarkers in the future.