CRII:SCH:A Generative Deep Learning (GDL) based Platform for Super-resolution, Virtual-Pathological Visualization of Coronary Images

冠状动脉疾病(CAD)已经影响了一个超过28亿美元的市场，每年大约进行100万次治疗。现有的CAD治疗指南空间分辨率有限，缺乏实时详细的病理识别。该项目旨在研究新的计算技术，用于冠状动脉图像的病理、超分辨率可视化。该项目如果成功，将有助于为心血管疾病的治疗提供新一代临床指导，心血管疾病目前是美国人类死亡的主要原因。从技术上讲，这种将多域图像表示编码成单域图像采集的方法，可以极大地降低硬件成本和医疗人力，从而有利于其他领域，如多摄像机监控、多模式生物医学成像等。该项目的教育计划强调旨在指导高级设计的活动，丰富深度学习课程的课程，并促进多元文化工程项目中少数族裔学生的外展。该项目旨在开发一种数据驱动的方法，利用离线数据和培训过程，在不进行任何硬件修改的情况下，产生以前无法获得或只能在体外获得的新信息。该项目将开发生成性深度学习算法，通过在脱机训练期间从高分辨率OCT图像和组织学显微图像中聚合图像信息，为低分辨率光学相干断层扫描(OCT)图像产生额外信息。该项目将研究如何通过体积生成性对抗网络(GAN)提高OCT的分辨率，同时保持快速扫描速度以实现超分辨率。该项目将开发一种新的非配对训练方案，通过使用基于GaN的图像翻译框架将OCT图像映射到组织病理学图像。该方法将通过对OCT图像和组织病理学图像的客观和主观分析来验证。该项目预计将在计算机科学和生物医学信息学方面产生学术成果。该项目将提供跨平台体积超分辨率的深度学习解决方案和解决跨通道图像翻译的生成性学习方法。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Coronary artery disease (CAD) has been influencing a market over $2.8 billion with roughly 1,000,000 treatment procedures performed annually. Existing guidance for CAD treatment suffers from limited spatial resolution and lacks real-time detailed pathological identification. This project is to investigate novel computational techniques for pathological, super-resolution visualization of coronary images. The project, if successful, will contribute towards a new generation of clinical guidance for the treatment of cardiovascular disease, which is currently the leading cause of human deaths in the United States. Technically, the method of encoding multi-domain image representations into a single-domain image acquisition could benefit other fields, such as multi-camera surveillance monitoring, multimodal biomedical imaging, etc., in terms of greatly reducing hardware cost and medical labor. The educational plan in this project emphasizes activities designed to guide senior designs, enrich curriculum in deep learning courses, and facilitate outreach for minority students in multicultural engineering programs.This project aims to develop a data-driven approach to use off-line data and training process, without any hardware modifications, to generatively produce new information that could not be acquired previously or can only be obtained ex vivo. This project will develop generative deep learning algorithms to produce additional information for low-resolution optical coherence tomography (OCT) images by aggregating image information from high-resolution OCT images and histological microscopic images during off-line training. This project will investigate on improving the resolution of OCT while maintaining fast scanning rate via a volumetric generative adversarial network (GAN) for super-resolution. This project will develop a novel unpaired training scheme to map OCT image to a histopathology image by using a GAN-based image translation framework. The approach will be validated using both objective and subjective analysis on OCT images and histopathology images. This project is expected to generate academic outcomes in both computer science and biomedical informatics. This project will provide a deep learning solution for cross-platform volumetric super-resolution and a generative learning approach to address cross-modality image translation.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.