Excellence in Research: PathoRadi ‒ an interactive web server for AI-assisted radiologic-pathologic image analysis, correlation and visualization

放射成像是医疗保健服务的重要组成部分，医生在医疗决策过程中严重依赖。现代放射学和成像科学的一个主要目标是利用专门的生物物理模型来模拟活组织中的生物过程，从而为疾病检测产生敏感的成像对比。为了了解模拟图像对比度与潜在病理生理之间的关系，必须进行放射-病理图像分析，以验证组织结构，病理和疾病特征中的图像相关性。鉴于组织中复杂的微环境，放射学和病理图像的比较尤其具有挑战性。实验室的许多常规分析在很大程度上依赖于人工或半自动计数和分割“金标准”病理图像中的细胞和组织，使用的是为通用目的而设计的市售软件。研究人员往往不得不放弃大量的信息，这些信息显示在病理图像中，但使用现有的方法无法量化。该项目旨在通过利用深度学习方法提取放射学和病理图像中的重要特征，以对以前在社区中无法实现的相关性进行定量分析，从而缩小差距。为了解决持续比较放射学和病理图像的挑战，该项目的技术目标分为三个方面：(1)用于定量全脑切片细胞形态表型的深度学习算法，(2)用于可视化放射学-病理图像相关分析的图形和交互式统计工具箱，(3)一个网站即服务软件包，在用户友好的平台上实现计算机辅助图像分析和放射学-病理相关性数据库。该项目成果提供了一种新的深度学习方法，可用于标准化放射成像生物标志物开发中的基准评估。该奖项通过在生物成像研究领域使用尖端人工智能，加强了霍华德大学研究生和本科生的STEM教育，为生物学和数学专业的多元化和代表性不足的学生提供了支持。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Radiological imaging is a critical part of healthcare services which physicians rely heavily upon in the medical decision-making process. A major goal of modern radiology and imaging sciences is to exploit specialized biophysical modeling that simulates the biological process in the living tissue to generate sensitive imaging contrast for disease detection. In order to understand the relations between the simulated image contrast and the underlying pathophysiology, radiologic-pathologic image analysis has to be performed to validate the image correlations in tissue structure, pathology and disease characteristics. Given the complex microenvironments in the tissues, comparison of radiologic and pathologic images is particularly challenging. Many of the routine analyses in the laboratories largely depend on manual or semi-automatic counting and segmentation of cells and tissues in the “gold standard” pathological images using commercially available software that are designed for general purposes. Researchers often have to give up an ample amount of information that shows in the pathological images but not quantifiable using the existing methods. This project aims to close the gap by utilizing deep learning methodology to extract the important features in the radiological and pathological images for quantitative analysis of the correlations previously unattainable in the community. To address the challenges that persist in comparing radiologic and pathologic images, the technical aims of the project are divided into three aspects: (1) deep learning algorithms for quantifying cell morphological phenotypes in the whole brain sections, (2) a graphical and interactive statistic toolbox to visualize the radiologic-pathologic image correlation analysis, (3) a website-as-a-service software package that implements computer-aided image analysis and database for radiologic-pathologic correlations in a user-friendly platform. The project outcome provides a novel deep learning methodology that can be used to standardize the benchmark evaluations in the development of radiological imaging biomarkers. The award enhances the graduate and undergraduate STEM education at the Howard University, with supports to a diverse and underrepresented cohort of the students in the biology and mathematics majors, through the use of cutting-edge artificial intelligence in the field of bioimaging research.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.