CAREER: Computational structured light imaging for widefield mapping of histologic primitives

这个CAREER项目的研究目标是创造新的光学和计算工具来改善未处理组织的组织学成像。组织学在大多数癌症的诊断和治疗中是必不可少的，但其核心工作流程在一个多世纪以来没有改变。本研究探索了紫外光模式和人工智能方法在大面积上直接成像组织学信息的潜力。这种“无滑动”的方法可以提高诊断的速度和准确性，同时改善手术和对各种癌症的评估。该研究计划与一个教育计划紧密结合，该计划旨在培养本科工程师解决病理学方面的重要临床需求，通过大型黑客马拉松将病理学家与工程师联系起来，并向高中生介绍医疗设备设计。传统的空间频域成像（SFDI）可以提供大组织的定量光学性质图，但由于调制深度随空间频率的增加而降低，导致分辨率和光学切片较差。该项目将创建一种新颖的宏观望远镜，该望远镜以非常高的空间频率投射紫外线照明，以恢复大块组织的高分辨率、表面光学特性图。将通过紫外激发（MUSE）显微镜测量量化细胞核形态、细胞结构和其他临床相关特征的组织学原始图谱。将获得离体人体皮肤样本的成对光学特性和组织学原始图，用于训练和测试机器学习模型。将研究新的多模态深度学习架构，以直接从宏输入预测这些组织原语，并联合优化宏硬件和预测算法。从这项研究中获得的知识对于1)弥合漫射光学成像和显微镜之间的知识差距，2)了解如何与成像硬件共同设计人工智能模型，以及3)为创新组织学工具奠定基础，这些工具可以显着改善病理工作流程。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。

The research goal of this CAREER project is to create new optical and computational tools to improve histological imaging of unprocessed tissues. Histology is essential in the diagnosis and treatment of most cancers, yet its core workflow has not changed in over a century. This research explores the potential of a patterned ultraviolet light and artificial intelligence approach to directly image histological information over large areas. This “slide-free” approach could increase speed and accuracy of diagnosis while improving surgery and assessment for a wide range of cancers. This research plan is closely integrated with an educational program that aims to train undergraduate engineers to solve important clinical needs in pathology, connect pathologists with engineers through large hackathons, and introduce high school students to medical device design.Conventional spatial frequency domain imaging (SFDI) provides quantitative optical property maps of large tissues but suffers from poor resolution and optical sectioning due to decreasing modulation depths with increasing spatial frequencies. This program will create a novel macroscope that projects ultraviolet illumination at very high spatial frequencies to recover high-resolution, superficial optical property maps of bulk tissues. Histologic primitive maps that quantify nuclei morphology, cellularity, and other clinically relevant features, will be measured via Microscopy by Ultraviolet Excitation (MUSE). Paired optical property and histologic primitive maps of ex-vivo human skin samples will be acquired for training and testing machine learning models. New multimodal deep learning architectures will be researched to predict these histologic primitives directly from macroscopy inputs and also to jointly optimize the macroscope hardware and prediction algorithm. The knowledge gained from this research will be important for 1) bridging the knowledge gap between diffuse optical imaging and microscopy, 2) understanding how to co-design artificial intelligence models with imaging hardware, and 3) creating a foundation for innovative histology tools that could dramatically improve pathology workflows.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.