The Neuropathologic Landscape of Alzheimer's Disease in Hispanic Decedents

摘要 公民科学，即公众参与科学过程，利用了人类认知的独特属性 将海量数据集的分析分发给普通公众。最近，人类计算研究所 (HCI)使用这种方法众包体内双光子激发显微镜(2PEF)数据的分析， 它将康奈尔大学的阿尔茨海默病(AD)研究加速了五倍，阐明了 阿尔茨海默病小鼠模型毛细血管停滞的分子机制。对于这份行政副刊， 我们现在建议开发和验证一个新的在线平台，该平台可以众包组织病理学分析 人类大脑的整个幻灯片图像(WSI)。这将支持快速获取大容量数据集 需要培训自动分类器以支持我们现有的研究神经病理的R01研究 西班牙裔和非西班牙裔白人后裔中的AD景观。研究级分类的推演 来自多源WSI的标签需要超过当今最好的自动化方法的数据质量要求。 因此，我们打算开发一种大众动力的分析管道，将组织病理学分析减少到 可由一般事务的非专家成员完成的一系列可管理的在线任务 公共的，以及将许多个体答案聚合成一个类似专家的单一答案的“群体智慧”方法 回答。为了实现这些目标，我们的具体目标是开发流程、接口和聚合 支持大众支持的分析渠道并评估其内部有效性所需的方法。我们将聘用 一种重复的交叉验证方案，以测试基于人群的分析将导致至少 相对于黄金，两种主要病理类型(斑块和缠结)的敏感性和特异性分别为95%和95% 由专家注释员提供的标准数据。如果这些验证研究成功，一个在线平台 结合这一分析管道将用于执行有史以来第一次阿尔茨海默氏症的从头分析 由一群志愿者(面向西班牙裔血统的人)制作的组织病理学图像，可用于培训 用于自动分析的神经网络模型，并将支持不断改进的可重复使用的数字 对不断扩大的病理目标曲目敏感的病理工作流工具。

ABSTRACT Citizen science, public participation in the scientific process, leverages the unique properties of human cognition to distribute the analysis of massive data sets to the general public. Recently, the Human Computation Institute (HCI) used this approach to crowdsource the analysis of in-vivo two-photon excitation microscopy (2PEF) data, which has accelerated Cornell University’s Alzheimer’s disease (AD) research by a factor of five, elucidating molecular mechanisms that underlie capillary stalling in mouse models of AD. For this administrative supplement, we now propose to develop and validate a new online platform that crowdsources the histopathological analysis of whole slide images (WSI) of the human brain. This would support the rapid acquisition of high-volume datasets needed to train an automated classifier in support of our existing R01 studies investigating the neuropathological landscape of AD within Hispanic and non-Hispanic White decedents. Deriving research-grade classification labels from multi-source WSI entails data quality requirements that exceed today’s best automated methods. Thus, we intend to develop a crowd-powered analytic pipeline that reduces the histopathological analysis into a step-wise series of manageable online tasks that could be accomplished by non-expert members of the general public, as well as “wisdom of crowd” methods for aggregating many individual answers into a single expert-like answer. In service of these objectives, our specific aims are to develop the process, interfaces, and aggregation methods needed to support the crowd-powered analytic pipeline and evaluate its internal validity. We will employ a repeated cross-validation protocol to test the hypothesis that the crowd-based analysis will result in at least 95% sensitivity and 95% specificity for two major pathology categories (plaques and tangles) relative to gold standard data provided by expert annotators. If these validation studies are successful, an online platform incorporating this analytic pipeline will be used to execute the first ever de novo analysis of Alzheimer’s histopathology imagery by a crowd of volunteers (geared towards those of Hispanic heritage), be used to train a neural network model for automating the analysis, and will support the ongoing improvement of a reusable digital pathology workflow tool that is sensitive to an expanding repertoire of pathology targets.