KULMAP: Human Kidney, urinary tract and lung mapping center

摘要 现代医学已进入大数据革命的早期阶段;显微镜、数字化 电子学、光子学和测序技术，以及放射学等现有技术， 能够从人类受试者大量生成多模式、多尺度、多组学数据。此外，本发明还 结果数据的数字化，加上计算机硬件和软件的先进状态， 为计算图像科学家提供了新的机会， 这些生物标志物来自大数据，其发现通过手动手段是难以解决的。两项重要努力 沿着这一方向是由美国国立卫生研究院精心策划的，即人类生物分子图谱 项目（HuBMAP）和肾脏精准医学项目（KPMP）。前者侧重于定义引用 不同组织的生物学尺度的解剖图谱。后者仅侧重于定义结构和 稳态肾脏的功能图谱及其疾病状态偏差。上述财团 致力于生成多尺度的组学数据。主要目标是融合大量的多模态 数据来开发参考和疾病患者中组织异质性程度的综合模型， 从而使组织的临床解释更加客观。在接近多尺度的崇高目标之前， 多模态数据融合，第一步是使用来自单一模态和单一模态的数据进行试点实验。 机构，以验证统计参考范围可以充分定义。假设成功，这条管道 可以扩展到不同的器官类型和规模。为了实现这一目标，在本HubMAP补充文件中， 应用，我们建议调查形态结构的多样性，在'参考'肾组织明场 使用全景卷积神经网络从HuBMAP、KPMP和其他来源获取完整的载玻片图像。我们 将比较所得结构分布异质性与从等效慢性肾脏获得的异质性 KPMP中的疾病病例，用于基准测试，以精确确定参考上限 结构分布PI是计算肾脏病理学方面的专家， 肾室客观量化结果、肾脏疾病的计算分类，以及 从肾组织图像计算预测临床生物统计学。此外，PI是研究者的一部分 KPMP团队，为KPMP数据的技术数据分析管道的开发做出贡献。主要研究者有 在分析GTEx肾组织图像数据以及构建基于云的Web浏览器方面拥有丰富的经验 可访问的图像和组学数据归档和可视化系统，内置插件，用于大规模人工智能分析 按比例缩放图像和组学数据，最终用户只需最低限度的技术知识即可操作。这项工作 将为HuBMAP提供一个全面的数字病理学框架，用于结构的图像分析， 在结构注释中，“参考”统计组织图谱生成，以及辅助病理学家进行大 体积注释和促进分子整合研究。

Abstract Modern medicine has entered the early phase of big data revolution; massive progress in microscopy, digital electronics, photonics, and sequencing technologies, in addition to established techniques like radiology, have enabled generation of multi-modal, multi-scale, multi-omics data in large volume from human subjects. Further, the digitization of the resulting data, coupled with the advanced state of computer hardware and software, has opened up new opportunities for computational image scientists to identify previously unknown statistical biomarkers from big-data whose discovery is otherwise intractable by manual means. Two important efforts along this direction are orchestrated by the National Institutes of Health; namely, the Human BioMolecular Atlas Program (HuBMAP) and Kidney Precision Medicine Project (KPMP). The former focuses on defining a reference anatomical atlas across biological scale for diverse tissues. The latter focuses solely on defining a structural and functional atlas of the homeostatic kidney and their disease state deviations. The consortiums mentioned above are committed to generate multi-scale, -omics data. The primary objective is to fuse the massive multi-modal data to develop a comprehensive model of the extent of tissue heterogeneity in reference and disease patients, so that clinical interpretations of tissue can be more objectified. Before approaching the lofty goal of multi-scale, multi-modal data fusion, the first step is to conduct pilot experiments using data from single modalities and single organs to validate that a statistical reference range can be adequately defined. Presuming success, this pipeline can then be scaled to diverse organ types and scales. Toward that objective, in this HubMAP supplemental application, we propose to investigate morphological structural diversity in ‘reference’ kidney tissue brightfield whole slide images from HuBMAP, KPMP, and other sources using a panoptic convolutional neural network. We will compare the resulting structural distribution heterogeneity with that obtained from equivalent chronic kidney disease cases from KPMP for benchmarking purpose to precisely establish the upper limit on the reference structural distributions. The PI is an expert in computational renal pathology, and has generated numerous results on objective quantification of renal compartments, computational classification of renal diseases, and computational prediction of clinical biometrics from renal tissue images. Further, the PI is part of the investigator team of KPMP, contributing to the development of a technical data analysis pipeline for KPMP data. The PI has significant experience in analyzing GTEx renal tissue image data, as well as building a cloud-based, web browser accessible image and omics data archival and visualization system with built-in plugins for AI analysis on large scale image and omics data, which requires minimal technical knowledge to operate by end-users. This work will provide HuBMAP a comprehensive digital pathology framework for image analysis of structures, efficiency in structural annotation, a ‘reference’ statistical tissue atlas generation, as well as assist pathologists with large volume annotations and facilitate molecular integration studies.