CAREER: Automated extraction of vessel data from images to construct new models for vascular networks in plants and animals

血液从心脏泵入毛细血管，通过一个高度分支和相互连接的网络，称为心血管系统，血管尺寸从微米到厘米不等，血流速度相差1000倍。在几乎所有的多细胞生物中，类似类型的维管网络对于资源的流动都是必不可少的，包括植物（木质部网络），昆虫（气管网络）和哺乳动物。了解哪些进化原则和环境因素驱动血管网络的结构，以及什么限制了它们的流动，可以帮助更好地理解器官的结构和功能，并对森林，食物网甚至肿瘤等不同系统产生影响。该研究项目将提供直接测试现有模型和开发新的更现实的模型所需的数据。输出将是用于从图像中提取血管数据的新软件（例如，磁共振成像（MRI），计算机断层扫描（CT）），一个大型的数据库，用于植物和动物的血管测量，识别特定分类群或组织共同的分支模式，以及新的理论。测量将包括血管半径、长度、分支比和分支角度。现有的模型与一些初步结果相矛盾，包括不对称分支（两个不同大小和不同流量的子血管）和自相似性（跨尺度重复出现的相似分支模式）的偏差。将构建新的模型，以纳入这些研究结果，并用于预测血管网络的几何形状和流量之间的连接。模型的一个关键预测是描述毛细管数量如何随网络体积变化的标度指数。为了使分支几何结构能够快速转化为流速预测，将根据一系列这些特征标度指数对血管网络进行分类。与这些研究目标紧密结合的是三个教育目标：1）教学生如何使用图像识别软件从生物领域的图像中提取数据，包括博物馆收藏和实验室，2）教学生如何将经验结果转化为方程，并测试具体的，机械的假设，和3）通过网站与公众和科学界分享数据，以建立大型，综合的数据库，这将加速科学研究和社区推广。高中生、本科生、研究生以及博士后研究人员都将在本研究期间接受培训。值得注意的是，该项目每年将为三名高中生提供暑期研究经验。这种培训的一个重要方面是教学生如何将联合收割机理论和经验数据相结合，以及如何通过出版物、网站、课程材料和专业会议上的演讲以及外联活动传播研究成果。PI拥有教育和培训各级学生的经验，并将积极招募代表性不足的群体的学生从事信息学研究。出版物、数据库、代码和课程材料都将通过网站提供。总之，这项工作将提供一个例子，说明计算机视觉技术如何用于从世界各地的生物图像中提取数据。

Blood is pumped from the heart to the capillaries through a highly branched and interconnected network known as the cardiovascular system, with vessel sizes that range from microns to centimeters and blood flow speeds that differ by a factor of 1000. Similar types of vascular networks are essential for the flow of resources in nearly all multicellular organisms, including plants (xylem networks), insects (tracheal networks), and mammals. Understanding which evolutionary principles and environmental factors drive the structure of vascular networks, and what constrains the flow through them, could help lead to a better understanding of organismic structure and function with implications for systems as diverse as forests, food webs, and even tumors. This research project will supply the data needed to directly test existing models and to develop new, more realistic models. Outputs will be new software for extracting vascular data from images (e.g., Magnetic Resonance Imaging (MRI), Computed Tomography (CT)), a large database for vascular measurements in plants and animals, identification of branching patterns common to specific taxa or tissues, and new theory. Measurements will include vessel radii, lengths, branching ratios, and branching angles. Existing models are contradicted by some preliminary results, including asymmetric branching (two daughter vessels of different sizes and different flow) and deviations from self-similarity (similar branching patterns recurring across scales). New models will be constructed to incorporate these findings and used to predict connections between the geometry of and flow through vascular networks. A key prediction of models will be scaling exponents that describe how the number of capillaries changes with network volume. In an attempt to enable quick translation of branching geometry into predictions for flow rate, techniques will be adapted to classify vascular networks according to a suite of these characteristic scaling exponents.Closely integrated with these research objectives are three educational goals: 1) teach students how to use image recognition software to extract data from images across biological fields, including museum collections and labs, 2) teach students how to translate empirical results into equations and test specific, mechanistic hypotheses, and 3) motivate data sharing with the public and scientific community via websites for large, comprehensive databases that will accelerate scientific research and community outreach. High school, undergraduate, and graduate students, as well as postdoctoral researchers will all be trained during this research. Notably, summer research experiences for three high school students will be provided for each year of the project. A critical aspect of this training is in teaching students how to combine theory and empirical data and how to disseminate research findings through publications, websites, curricular materials, and talks at professional meetings as well as outreach. The PI has experience educating and training students at all levels and will actively recruit students from under-represented groups to engage in informatics research. Publications, databases, code, and curricular materials will all be made available through websites. Together, this work will provide an example of how computer vision techniques can be used to extract data from biological images around the world.