Augmented serial blockface histology: Toward a better understanding of 3D tissue microstructure

在过去的几十年里，成像技术的进步帮助科学界将功能与大脑内部的结构联系起来。磁共振成像(MRI)和正电子发射断层扫描(PET)等神经成像手段彻底改变了我们对大脑的理解。这些工具已被用于研究大脑代谢和神经退行性疾病等研究。尽管核磁共振和正电子发射计算机断层成像在神经科学中的应用无处不在，但当涉及到微米尺度的大脑结构研究时，这些技术并不是很合适。在神经光子学领域，一种满足这种不断增加的分辨率要求的工具是连续块面组织学(SBH)。这项技术结合了组织切片设备和光学显微镜。样品被顺序切片，以揭示新的组织层，这些组织层用显微镜成像。重复这一过程，直到对整个样本进行成像。然后，通过先进的配准方法，将获取的数千个图像块组装成一个单一的3D体积。SBH是许多备受瞩目的神经科学项目的重要组成部分，绘制了全基因组的基因表达图，并在整个小鼠大脑中获得了微米级的连接体。主要挑战之一是SBH为每个大脑生成了海量数据。例如，以提供1微米采样分辨率的40倍目标采集整个小鼠大脑，使用当前的SBH系统估计需要60天的采集时间，并且需要大约700 TB的磁盘空间来存储原始数据集。这对数据管理、重建和分析方法提出了挑战。未来在神经光子学领域的应用将越来越需要成像和机器学习之间更紧密的协同。我的发现号研究计划有两个目标：(1)通过将显微镜与先进的计算机视觉方法相结合来加速图像采集和重建，以及(2)使用基于机器学习的方法分析由这种成像系统产生的大量数据。这项研究计划中提出的项目旨在通过将显微技术与新的计算机视觉方法、图像处理和机器学习技术相结合来创建增强的连续块面组织学系统。这将使这种成像技术更快、更智能、更具重复性。这将为SBH的民主化开辟道路，是生物医学研究和临床社区更广泛采用这项技术的必要步骤。

Over the last decades, technological advances in imaging assisted the scientific community for linking functions to structures inside the brain. Neuroimaging modalities such as magnetic resonance imaging (MRI) and positron emission tomography (PET) have revolutionized our understanding of the brain. These tools have been used to study, among other investigations, brain metabolism and neurodegenerative diseases. Despite the ubiquitous use of MRI and PET imaging in neuroscience, these techniques are not well suited when it comes to study the brain structure at a micrometer scale. In the field of neurophotonics, a tool that meets this increasing resolution requirement is serial blockface histology (SBH). This technology combines a tissue slicing apparatus with an optical microscope. The sample is sequentially sliced to reveal new tissue layers that are imaged with the microscope. The process is repeated until the whole sample has been imaged. Then, through advanced registration methods, the thousands of image tiles acquired are assembled into a single 3D volume. SBH was an essential component of many high-profile neuroscience projects mapping genome-wide gene expression and for obtaining a micrometre scale connectome in a whole mouse brain. One of the main challenges is that SBH generates a tremendous amount of data for every brain. For example, to acquire an entire mouse brain with a 40X objective offering sampling resolution of 1 µm would require an estimated acquisition time of 60 days with current SBH systems and would necessitate around 700 Terabytes (TB) of disk space to store the raw dataset. This represents a challenge for data management, reconstruction, and the analysis methods. Future applications in this neurophotonics field will increasingly require a closer synergy between imaging and machine learning. My Discovery research program has two objectives: (1) to accelerate image acquisition and reconstruction by integrating the microscope with advanced computer vision methods, and (2) to analyze the large amount of data generated by such imaging systems with machine learning based methods. The projects proposed in this research program aim to create an augmented serial blockface histology system by integrating the microscopy with novel computer vision methods, image processing and machine learning techniques. This will make this imaging technique faster, smarter and more reproducible. This will open the way to democratization of SBH and is a necessary step toward the broader adoption of this technology by the biomedical research and clinical communities.