面向超薄序列切片的三维电镜自动成像技术研究

The construction of full-scale brain network map is a high ground of brain science and neuroscience. The development of ultrathin serial sections technology and three-dimensional (3D) electron microscopy (SEM) imaging technology promotes the progress in analysis of micro-scale structure map of brain. the 3D imaging of ultrathin slice sequence is a imaging process that connect the macro profile to the micro details of the slices, it across slice sequence carrier at decimeter, slices at millimeter, microscopy FOV at micron, and to the region of the interest at nanometer. In the face of big data needs of micro scale neural circuit analysis, there are some key problems of current solutions in the acquisition process, such as the wrong area of imaging, the wrong focus of the image, the low efficiency of the imaging scheme, This project intends to carry out the original research on the key technologies of the automatic 3D imaging technology for the SEM as follow: (I) Study on multi-scale image features with online random forest model, to solve the high precision automatic positioning problem; (II) Design the rapid evaluation algorithm of image clarity, to solve the auto-focus problem at different slices; (III) Implement the modularization and parallelization of the automatic SEM imaging system, to improve the efficiency of automatic SEM imaging at system level. Relying on the existing SEM cluster and the full localized high throughput scanning electron microscope, the research results will apply to a high throughput 3D electron microscopy imaging platform, providing a solution for the large volume of neural structure analysis.

构建全尺度脑网络图谱是脑科学与神经科学研究的一个战略制高点；超薄序列切片技术和三维电镜技术不断发展，推动了脑微观结构解析技术的进步。超薄序列切片三维成像是一个宏观概貌与微观细节相联系的成像过程，同时跨越分米级切片载体、毫米级切片、微米级显微视场，直至纳米级感兴趣区域。面对微观尺度神经环路解析的大数据需求，现有采集方案在采集过程中存在拍错感兴趣区域、图像虚焦、成像方案效率低等几个关键问题；本项目拟针对超薄序列切片的三维电镜自动成像技术的关键问题，开展如下原创性研究：（一）图像多尺度复合特征在线随机森林模型研究，解决高精度自动定位问题；（二）图像清晰度快速评价算法设计，解决切片自动聚焦问题；（三）自动成像系统模块化和并行化实现，从系统层面解决成像效率问题。依托现有的电镜集群与完全国产化的高通量扫描电镜，研究成果将应用于高通量三维电镜自动成像平台，为大体量神经结构解析提供解决方案。

依托国家自然科学基金立项，为解决生物结构高通量扫描电镜三维影像系统的技术瓶颈，打破国外扫描电镜在微观重建领域的仪器相关技术垄断，促进我国生物电镜相关成像技术的发展，本课题针对生物序列高通量扫描电镜三维自动化成像的关键技术开展研究。.本课题研究解决的关键问题主要包括：（一）解决序列切片感兴趣区域高精度自动定位问题；（二）解决序列切片图像清晰成像问题；（三）实现自动成像系统模块化和并行化；（四）如何通过基于深度学习的超分辨率技术提高三维电镜成像效率的研究。.本课题支持的相关研究共发表论文3篇，其中SCI收录1篇，中文核心期刊2篇；提交中国发明专利3项，国际发明专利1项，其中已授权中国发明专利2项，国际发明专利1项（PCT美国授权），本课题负责人均为第一发明人。基于课题研究内容的自主研发扫描电镜成像系统两套，一套应用于德国蔡司接口可控的扫描电镜，取得软件著作版权；另一套在国产聚束科技扫描电镜上使用，目前已投入实际生产测试，尚在完善和整理中。.课题按照研究计划稳步执行，为国内相关科研单位获取大规模微观脑图谱数据，支撑了相关科研单位的科学项目研究，推进了国产电镜在生物序列扫描电镜成像技术产业化发展。