Biopsy-free, label-free 3D virtual histology of intact skin

在美国，每年诊断出超过200万种皮肤癌，其中80%是基底细胞癌。目前的护理标准仍然是侵入性活检，然后进行评估，导致不必要的疤痕，多次患者就诊，并增加了医疗保健系统的成本。在微观层面上对皮肤进行非侵入性成像可以立即对皮肤癌进行分类。该项目将开发一种基于机器学习的技术，将完整皮肤的3D图像转换为易于解释的高质量虚拟染色图像。该项目开发的技术可以让皮肤科医生获得准确和快速的皮肤病变诊断，并改善临床护理。此外，这项技术可以改善患者对皮肤科医生和皮肤科诊所工作流程的访问。它还可以在与患者互动有限的时间内提供远程皮肤病学解决方案。该项目还将推进一个补充教育和推广计划，其中将涉及公共采访和新闻媒体和互联网上的科普文章;在调查员实验室为代表性不足的本科生提供研究机会;通过组织研讨会，研讨会和会议对研究生进行培训。这些活动将帮助本科生和高中生与前沿的研究环境互动，提高他们的科学好奇心，并塑造他们在科学和工程领域的职业目标。该项目的总体目标是开发基于深度学习的数字染色技术，从根本上提高皮肤科医生准确诊断皮肤病变的能力，而不需要皮肤活检或任何组织的化学处理。基于深度神经网络的计算方法将用于训练虚拟组织学算法，以获得与正常皮肤和疑似基底细胞癌（最常见的皮肤癌类型）的肿瘤相对应的完整皮肤的准确、虚拟染色图像。此外，将使用各种测试来确定和量化虚拟组织学格式是否可以改善诊断。这些将包括与皮肤的实际组织学的比较，以及由接受过反射共聚焦显微镜培训的皮肤科医生进行的诊断准确性（灵敏度和特异性）的客观性能指标，与诊断的金标准：组织学处理进行比较。一旦成功，这些研究将提供一个跨越式的进步，朝着无活检皮肤组织学评估，这有可能改变临床皮肤病学的做法。这个奖项反映了NSF的法定使命，并已被认为是值得支持的评估使用基金会的知识价值和更广泛的影响审查标准。

Over 2 million skin cancers are diagnosed annually in the United States, with 80% being basal cell carcinoma. The current standard of care remains invasive biopsy followed by assessment, resulting in unnecessary scars, multiple patient visits, and added costs to the healthcare system. Non-invasive imaging of the skin at the microscopic level could allow for immediate triage of skin cancers. This project will develop a machine learning-based technique to transform 3D images of intact skin into high quality virtually stained images that can be easily interpreted. The technology developed in this project can allow dermatologists to obtain accurate and rapid diagnosis of skin lesions and improve clinical care. Furthermore, this technology may improve patient access to dermatologists and dermatology clinic workflow. It may also provide tele-dermatology solutions during times with limited patient interactions. This project will also advance a complementary education and outreach program which will involve public interviews and popular science articles in news media and internet; research opportunities in the investigator's laboratory for underrepresented undergraduate students; and graduate student training through the organization of workshops, seminars, and conferences. These activities will serve undergraduate and high school students to interact with a cutting-edge research environment, increasing their scientific curiosity and shaping their career goals in science and engineering.The overarching goal of this project is to develop deep learning-based digital staining technology that can fundamentally improve dermatologists’ ability to accurately diagnose skin lesions without requiring skin biopsy or any chemical processing of tissue. Deep neural network-based computational approaches will be used to train the virtual histology algorithm for obtaining accurate, virtually stained images of intact skin corresponding to normal skin and neoplasms suspicious for basal cell carcinoma, the most common type of skin cancer. Furthermore, various tests will be used to determine and quantify whether the virtual histology format can improve diagnosis. These will include comparisons with actual histology of the skin and objective performance measures of diagnostic accuracy (sensitivity and specificity) by dermatologists who are trained in reflectance confocal microscopy when compared to the gold standard for diagnosis: histological processing. Once successful, these studies will provide a leapfrog advancement towards biopsy-free histologic assessment of the skin, which has the potential to transform the practice of clinical dermatology.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.