Informatics for Integrative Brain Tumor Whole Slide Analysis

描述(由申请人提供)： 数字化病理切片的高分辨率图像分析与分子数据相结合，具有巨大的潜力，可以为患者在预后和治疗方面的分层提供额外的信息。我们建议开发方法、分析管道和数据管理工具，使系统地对脑瘤组织学特征以及蛋白质和基因表达模式进行大规模比较分析成为可能。我们将开发信息模型来管理与脑瘤全虚拟幻灯片数据分析相关的信息。这些模型将捕获与患者数据、标本准备和特殊染色、涉及组织学分类和特征的人类观察、算法组成、参数化和与分析管道对应的输入数据、以及算法和人类描述的分割、特征和分类相关的信息。我们将为高性能数据库和查询支持实现中间件，这些查询基于图像和来源信息的元数据选择图像数据和结果的子集；比较从多种算法获得的特征、空间结构和分类以及人工标记；以及比较多个图像数据集上的特征和分类的统计和摘要信息。利用信息模型和中间件，我们将开展必要的分析研究，以确定图像分析衍生的肿瘤信息与临床结果、基因表达类别、遗传得失和甲基化状态之间的关系。我们将采用一种新的具有肽控制和定量图像分析方法的自动多重量子点免疫组织化学方法，使用组织学特征描述来绘制相对于肿瘤微环境的信号转导途径和转录网络的活动。我们将利用多变量数据融合技术，同时考虑测量的图像特征、分子特征之间的潜在相关性和关系，以预测患者的结果。 我们将部署一个数据存储库，其中包含来自我们项目的图像、特征、分析管道、来源信息和分析结果。这个资料库将为脑瘤研究提供一个公开可用的资源。在这个项目中开发的所有软件和信息模型都将是开源的，并免费供研究使用。

DESCRIPTION (provided by applicant): High-resolution image analysis of digitized pathology slides coupled with molecular data has enormous potential to provide additional information for stratifying patients in terms of prognosis and therapy. We propose to develop methods, analytic pipelines, and data management tools that will make it feasible to systematically carry out large-scale comparative analyses of brain tumor histological features and of patterns of protein and gene expression. We will develop information models to manage information associated with analysis of brain tumor whole virtual slide data. These models will capture information about context relating to patient data, specimen preparation, and special stains, human observations involving histological classification and characteristics, algorithmic composition, parameterization and input data corresponding to analysis pipelines, and algorithm and human-described segmentations, features, and classifications. We will implement middleware for high-performance database and query support for queries that selects subsets of image data and results based on metadata on images and provenance information; that compare features, spatial structures, and classifications obtained from multiple algorithms as well as human markups; and that compare statistical and summary information on features and classifications across multiple image datasets. Using the information models and middleware, we will carry out analysis studies needed to determine the relationship between image analysis derived tumor information and clinical outcome, gene expression category, genetic gains and losses, and methylation status. We will employ a novel automated multiplex quantum dot immunohistochemistry with peptide controls and quantitative image analysis methodology to map the activity of signal transduction pathways and transcriptional networks relative to the tumor microenvironment using histology feature descriptions. We will leverage multivariate data fusion techniques to simultaneously take into account potential correlations and relationships among the measured image features, molecular signatures to predict patient outcomes. We will deploy a data repository populated with images, features, analysis pipelines, provenance information, and analytic results from our project. This repository will provide a publicly available resource for brain tumor research. All software and information models developed in this project will be open source and free for research use.