中国科学院自动化研究所机构知识库

理解大脑工作机制对基础脑神经科学和脑疾病的研究具有重要意义。复原脑组织结构，并对其进行识别和分析是揭示脑工作原理的重要途径。纳米尺度的脑微观结构重建可以有效复原脑组织的三维结构，为后续研究提供数据基础。目前，随着电子显微成像技术在生命科学研究中的深入应用，纳米尺度的微观重建技术正向着更高的空间分辨率和更大的重建体积发展。作为脑微观重建的重要环节，序列切片电镜图像拼接配准也成为限制大规模脑微观结构重建的瓶颈。所以开展相关算法研究，提升拼接配准精度，优化拼接配准流程对脑微观结构重建至关重要。

本文是脑微观结构三维电镜重建研究的一部分，探究脑微观结构拼接配准中的线性形变矫正方法。通过构建不同的线性形变矫正模型，估计和矫正电镜切片和拍摄等过程中产生的线性形变，实现脑组织切片图像的拼接和配准。

依照脑微观结构重建流程，本文的研究分两步进行：首先在2D方向上，对大范围切片电镜图像进行拼接。通过对扫描电镜成像过程中产生的畸变进行估计，避免了图像边缘畸变对拼接结果造成的影响。并通过整体优化的方式矫正成像过程中引入的线性形变，得到了无缝的拼接结果。实验结果表明，该方法能够达到亚像素的拼接精度，有效的克服了电镜高分辨率下视场的局限性，扩大了重建范围；在3D方向上，优化现有的仿射配准流程，克服配准流程不合理带来的误差累积和传播。完成了序列切片图像的粗尺度配准，矫正切片图像位于切面内的线性形变并尽可能的保留了图像原始的结构信息，恢复了脑组织结构在z方向的连续性。结果显示本文提出的拼接配准方法在实际的大体量脑微观结构重建中更具适用性。

本文旨在设计高效的序列切片电镜图像线性形变矫正方法，实现切片图像的拼接配准，提升脑微观结构拼接配准的精度，有效的复原出大范围脑组织微观结构的三维形貌，为后续的研究提供良好的数据基础。

Understanding the working mechanism of the brain is of great significance to the research of basic brain neuroscience and brain diseases. Recovering brain tissue structure and analyze it is one of the important methods to reveal the working principle of the brain. Nanoscale brain microstructure reconstruction can effectively restore the three-dimensional structure of brain tissue, and provide data basis for the following research. At present, with the application of electron microscopic imaging technology in bioscience research, Nanoscale reconstruction technology is developing towards higher spatial resolution and larger reconstruction volume. As an important part of brain microstructural reconstruction, image stitching and registration of serial section electron microscope have become the bottleneck of large-scale brain microstructural reconstruction. Therefore, it is very important to research relevant algorithm to improve the stitching and registration accuracy and optimize the registration process for brain microstructure reconstruction.

This paper is a part of the study of brain microstructural reconstruction and explores the linear deformation correction method in brain microstructure stitching and registration. It has constructed different linear deformation correction models to estimate and correct the linear deformation in the process of sectioning and imaging and achieves the purpose of brain tissue slice image mosaic and registration.

According to the procedure of brain microstructure reconstruction, the study is divided into two steps: Firstly, a large region of section electron microscope images is stitched in a two-dimensional direction. By estimating the distortion introduced in the process of scanning electron microscope imaging, the negative effect of image edge distortion on the stitching results is avoided. And the linear deformation introduced in the imaging process is corrected by overall optimization, and the seamless stitching result is obtained. Experimental results show that this stitching method can achieve sub-pixel accuracy, and overcome the limitation of the high-resolution field of view of electron microscope effectively. In a three-dimensional direction, the affine registration process is optimized to overcome the error accumulation and propagation caused by the unreasonable registration process. So, the coarse-scale registration of serial images is completed, the linear deformation of slice images in the section plane is corrected, and the original structure information of the image is retained as much as possible. The results show that the stitching and registration method proposed in this paper is more suitable for the actual reconstruction of large brain microstructure.

The purpose of this paper is to design efficient linear deformation correction methods for serial section electron microscope images, improve the accuracy of brain microstructure stitching and registration, restore the large region of three-dimensional morphology of brain microstructure effectively, and provide a good data basis for the following research.