CAREER: Microscopy Image Analysis to Aid Biological Discovery: Optics, Algorithms, and Community

该项目开发图像分析算法和系统来处理显微镜图像，记录生物标本的增殖历史，并评估它们对不同培养条件的反应，从而破译复杂的生物过程，加速生物发现的进展。该研究结合了物理光学、计算机视觉和众包技术，为显微镜成像和显微镜图像分析带来了突破。这些技术的发展使基于图像的生物学研究从主观性转变为严谨、定量和高效的方式。研究团队亦致力推动生物影像与电脑视觉的跨学科合作，将研究成果整合到教育活动中，并透过网络、K-12小组、会议及业界合作等方式，将项目传播给更广泛的受众。以往的显微图像分析方法没有考虑特定的图像形成过程，将其与一般的自然图像等同对待，造成了图像分析的许多困难或失败。该项目通过研究显微镜光学的理论基础，以一种主要不同的方式解决了这些挑战。导出了显微镜的计算成像模型，并将其用于恢复无伪影图像和提取面向光学的图像特征，使自动图像分析从根本上正确易行。这些模型进一步用于增强显微镜的功能，包括校准和虚拟显微镜。正在建立一个网络研究社区，利用主动学习和群体计算来提高算法性能和生物发现。更新可从http://web.mst.edu/~yinz/获得。

This project develops image analysis algorithms and systems to process microscopy images that record the proliferation history of biological specimens and evaluate their behaviors that respond to different culturing conditions, therefore, deciphering complex biological processes and accelerating the advance of biological discovery. The research combines techniques of physical optics, computer vision and crowdsourcing to bring a breakthrough to microscopy imaging and microscopy image analysis. The developments of such technologies transform the image-based biology research from subjective to a rigorous, quantitative, and efficient manner. The research team also seeks to promote interdisciplinary collaboration between biological imaging and computer vision, integrate the research outcomes into education activities, and disseminate the project to a wide audience via web, K-12 group, conferences and industry collaborations.Previous microscopy image analysis methods do not consider the particular image formation process and treat them in the same manner as general natural images, causing many difficulties or failures in the image analysis. This project addresses the challenges in a principally different way by investigating the theoretical foundation of microscopy optics. The computational imaging models of microscopes are derived and used to restore artifact-free images and extract optics-oriented image features, which makes the automated image analysis fundamentally correct and easy. The models are further used to enhance the microscope's functionalities including calibration and virtual microscopy. A cyber-enabled research community is being established within which active learning and crowd-computing are leveraged to improve the algorithm performance and biological discovery.Updates are available from http://web.mst.edu/~yinz/.