CAREER: Developing Algorithms for Object-Adaptive Super-Resolution in Biomedical Imaging

先进的生物医学成像技术通过提供结构和功能细节，彻底改变了诊断和治疗。然而，由于图像采集时间的限制，生物医学图像的空间分辨率有时不能满足特定的应用。传统的基于软件的改进在计算和可视化方面的成本很高，这为以合理的成本将框架优化到超分辨率打开了一个利基。它与NSF的使命一致，即促进计算机科学的进程和推进国民健康。本课题旨在研究一种新的算法，以自适应地提高数字分辨率，并将超分辨率过程中的计算成本降至最低。从技术上讲，该方法结合了目标检测和超分辨率的努力，带来了一个通用的工具，潜在地有利于多种生物医学成像模式，如光学相干断层扫描（OCT），组织学显微镜，共聚焦图像，MRI，超声等。教育强调活动，以扩大代表性不足的群体在生物医学追求的参与。本项目旨在开发智能对象自适应超分辨率算法，以鲁棒、高效和可推广的方式提高生物医学图像的分辨率。该项目将开发鲁棒的目标检测神经网络来识别需要超分辨的区域。确定自适应超分辨率的尺度因子。本项目将研究在复杂值图像重建过程中超分辨生物医学图像到多个尺度因子的高效计算算法。该项目还将开发一个可转让的框架，以便在一种成像模式下开发的超分辨率技术可以适应于由不同成像模式开发的超分辨率技术。这些方法将使用OCT数据进行验证，并且将通过从OCT域转移到组织病理学域来验证域适应。研究成果还将产生基于人工智能的教育材料和软件，以减少对生物医学设施的需求，这些设施通常是必需的，但对于代表性不足的群体来说成本效益不高。此外，该项目还包括外展活动，以促进生物医学资源有限地区的生物医学参与，并建立一个新模式，指导一个多样化和包容性的团队，并为下一代研究人员创造动力。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。

Advanced biomedical imaging technology has revolutionized diagnosis and treatment by providing structural and functional details. Spatial resolution of biomedical images, however, sometimes do not suffice for specific applications due to constraints of image acquisition time. Conventional software-based improvement bears high costs in computation and visualization, opening a niche to optimize the framework towards super-resolution at reasonable costs. It aligns with NSF’s mission to promote the process of computer science and to advance the national health. This project is to investigate novel algorithm development to adaptively improve digital resolution and minimize the cost of computation in super-resolution process. Technically, this method combines the effort of object detection and super-resolution to bring a generalizable tool to potentially benefit multiple biomedical imaging modalities, such as optical coherence tomography (OCT), histological microscopy, confocal images, MRI, ultrasound, etc. The educational emphasizes activities to broaden the participation of underrepresented groups in biomedical pursuits.This project aims to develop intelligent object-adaptive super-resolution algorithms to improve resolutions of biomedical images in a robust, efficient, and generalizable manner. This project will develop robust object detection neural network to identify regions to be super-resolved. A scale factor will be determined for adaptive super-resolution. This project will investigate on computationally efficient algorithms to super-resolve biomedical images to multiple scale factors during a complex-valued image reconstruction process. This project will also develop a transferrable framework such that the super-resolution technology developed in one image modality can be adapted into the super-resolution technology developed by a different imaging modality. The approaches will be validated using OCT data and the domain adaption will be validated by transferring from OCT domain to histopathological domain. The research outcome will also result in artificial intelligence-based educational materials and software to reduce the need of biomedical facilities that are conventionally required but not cost-effective to underrepresented groups. In addition, this project includes outreach activities to promote biomedical participation in regions with limited access to biomedical resources and a new model to mentor a diverse and inclusive team and create motivation to the next generation of researchers.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.