Developing computational algorithms for histopathological image analysis

项目摘要 组织病理学是疾病诊断和预后的基石。随着成像技术的进步 技术，组织切片的全幻灯片图像(WSI)扫描正在成为一种常规的临床程序和 产生海量数据，以高分辨率捕捉组织病理学细节。最新版本 病理图像分析方法类似于一般的图像分析方法，主要侧重于形态学 特征，如组织的纹理和粒度，但忽略了组织的复杂的等级结构。单元格 是构成组织的基本构件。不同类型的细胞首先被组织成细胞 这些成分与细胞外基质一起形成了不同类型的组织结构。 了解这些不同类型的细胞之间的相互作用可以提供对生物学和 疾病状况。然而，存在一些主要的计算挑战：(1)如何识别和分类 组织中不同类型的细胞，(2)如何表征高度复杂和异质的空间 组织的组织，以及(3)如何将组织病理学数据与其他类型的数据相结合来研究疾病 现状和进展。这项提议的目标是开发新的计算方法来分析 用于研究疾病状态和进展的组织病理学图像数据。为了实现这一目标，我们已经建立了 一个强大的研究团队，在图像分析、机器学习、统计建模、 和临床病理学。具体地说，我们将开发新的算法来：(1)将不同类型的细胞从 组织病理学组织WSI扫描，(2)表征和量化细胞空间分布和细胞-细胞相互作用， (3)将组织病理学数据与其他类型的数据相结合，以研究疾病进展。所有建议的方法 受到不同类型疾病(如肝脏)的实际生物和临床应用的推动 疾病、传染病和癌症。如果实施成功，拟议的研究将有助于 对从组织病理学组织载片产生的数据进行分析和建模以改进疾病风险评估， 诊断和结果预测。

Project Summary Histopathology is the cornerstone of disease diagnosis and prognosis. With the advance of imaging technology, whole-slide image (WSI) scanning of tissue slides is becoming a routine clinical procedure and producing a massive amount of data that captures histopathological details in high resolution. Most current pathological image analysis methods, similar to general image analysis approaches, mainly focus on morphology features, such as tissue texture and granularity, but ignore the complex hierarchical structures of tissues. Cells are the fundamental building blocks to tissues. Different types of cells are first organized into cellular components, which together with the extracellular matrix, form different types of tissue architectures. Understanding the interactions among these different types of cells can provide critical insights into biology and disease status. However, there are some major computational challenges: (1) How to identify and classify different types of cells in tissue, (2) how to characterize the highly complex and heterogeneous spatial organization of tissue, and (3) how to integrate histopathology data with other types of data to study disease status and progression. The goal of this proposal is to develop novel computational methods to analyze histopathology image data to study disease status and progression. In order to achieve this goal, we have built a strong research team with complementary expertise in image analysis, machine learning, statistical modeling, and clinical pathology. Specifically, we will develop novel algorithms to: (1) classify different types of cells from histopathology tissue WSI scans, (2) characterize and quantify cell spatial distribution and cell-cell interactions, and (3) integrate histopathology data with other types data to study disease progression. All proposed methods were motivated by real-world biological and clinical applications across different types of diseases, such as liver diseases, infectious diseases, and cancer. If implemented successfully, the proposed study will facilitate the analysis and modeling of data generated from histopathology tissue slides to improve disease risk assessment, diagnosis, and outcome prediction.