毕业生知识库

  理解脑的神经机理对基础神经科学、神经疾病研究、类脑智能探索具有重要意义，借助电子显微镜重建突触水平的脑微观连接是开展神经科学相关研究的重要手段。随着脑科学的发展，相关研究需要更大规模、更高精度的重建数据，传统的体电镜成像技术逐渐无法满足以上成像需求。为此，中国科学院自动化研究所发展出了更先进的连续切片刻蚀减薄技术，该技术能够对大范围的生物切片进行高精度的减薄，为体电镜提供了大规模、高精度的生物样片。由于基于该技术的体电镜成像方案尚未成熟，亟待开展相关的成像技术研究，优化面向连续切片刻蚀减薄的电镜成像流程，推动大规模、高精度脑微观重建工作。

  本文围绕连续切片刻蚀减薄技术，对国际上相关的体电镜成像技术进行了深入的研究，针对该连续切片刻蚀减薄技术特点，设计并实施了可行的成像流程、相关算法和三维成像系统，实现了大规模生物组织的高精度三维成像。具体地，本研究解决了刻蚀过程中切片厚度动态变化、连续切片多次定位成像带来的相关问题。由于切片机的随机误差，连续切片的厚度具有一定差异，不同切片的实际刻蚀次数可能不同，为了通过厚度信息判断实际刻蚀次数，本文提出了一种针对刻蚀切片的厚度估计算法。该算法通过卷积神经网络计算生物结构形变，使用形变信息度量切片差异，避免了成像衬度差异造成的度量偏差，通过构建带有厚度真值的电镜图像，建立了形变与厚度之间的估计模型。实验结果表明，该方法提高了切片厚度的估计精度，并对训练集中不存在的厚度数据具有很好的泛化性能。针对连续切片刻蚀减薄的大规模电镜成像任务，本文分析了连续切片刻蚀减薄电镜成像相较于连续切片电镜成像的不同，并提出了面向连续切片刻蚀减薄的电镜成像流程和成像系统，该流程在连续切片电镜成像流程的基础上增加了切片厚度判断、感兴趣区域重定位环节，本文构建的面向连续切片刻蚀减薄电镜成像系统能够通过简便的人工交互实现半自动感兴趣区域定位、成像、晶圆重载等功能。实际应用表明，该成像系统能够精准、高效的完成面向连续切片刻蚀减薄的微观三维成像任务。

  本文旨在设计一套针对连续切片刻蚀减薄技术的完整体电镜成像方案，构建高质量、高精度、高效率的三维成像系统，通过计算的方式提升成像系统的自动化程度，为连续切片刻蚀减薄成像方案建立良好的技术基础。

Understanding the neural mechanisms of the brain is of great significance for basic neuroscience, neurological disease research, and exploration of brain-like intelligence. The use of electron microscope to reconstruct synaptic level brain micro connections is an important means of conducting neuroscience related research. With the development of neuroscience, related research requires larger scale and higher precision reconstruction data, and traditional volume electron microscopic imaging techniques are gradually unable to meet the above imaging needs. For this reason, the Institute of Automation of the Chinese Academy of Sciences has developed a more advanced etching thinning continuous sections technology, which can achieve high-precision thinning of biological sections in a large area and provide large-scale and high-precision biological samples for the volume electron microscope. Due to the immaturity of the volume electron microscopic imaging scheme based on this technology, it is urgent to carry out relevant imaging technology research, optimize the electron microscopic imaging process for etching thinning continuous sections, and promote large-scale and high-precision brain micro reconstruction work.

This thesis focuses on the etching thinning continuous sections technology and conducts in-depth research on relevant volume electron microscopic imaging technologies internationally. Based on the characteristics of this etching thinning continuous sections technology, feasible imaging processes, related algorithms, and 3D imaging software and hardware systems are designed and implemented, achieving high-precision 3D imaging of large-scale biological tissues. Specifically, this study addresses the issues related to dynamic changes in section thickness during etching and multiple localization and imaging of continuous sections. Due to the random error of the microtome, the thickness of continuous sections varies to a certain extent, and the actual etching times of different sections may vary. To determine the etching times based on thickness information, this thesis proposes a thickness estimation algorithm for etching sections. This algorithm calculates biological structural deformation through convolutional neural networks and uses deformation information to measure section differences, avoiding measurement bias caused by imaging contrast differences. By constructing electron microscope images with thickness truth values, an estimation model between deformation and thickness is established. The experimental results show that this method improves the estimation accuracy of section thickness and has good generalization performance for thickness data that does not exist in the training set. For the large-scale electron microscopy imaging task of etching thinning continuous sections, this thesis analyzes the differences between etching thinning continuous sections electron microscopic imaging compared to continuous sections electron microscopic imaging, and proposes an electron microscopic imaging process and imaging system for etching thinning continuous sections. This process adds section thickness judgment and region of interest repositioning on the basis of the continuous sections electron microscopy imaging process. The electron microscopic imaging system for etching thinning continuous sections constructed in this thesis can achieve semi-automatic region of interest positioning, imaging, wafer reloading and other functions through simple manual interaction. Practical applications have shown that the imaging system can accurately and efficiently complete micro 3D imaging tasks for etching thinning continuous sections.

This thesis aims to design a complete volume electron microscopic imaging scheme for etching thinning continuous sections technology, construct a high-quality, high-precision, and high-efficiency 3D imaging system, improve the automation level of the imaging system through calculation, and establish a good technical foundation for etching thinning continuous sections imaging scheme.