Improvements in semiautomated serial section reconstruction and visualization of neural tissue from TEM images

摘要

Abstract: Our initial system for 3D reconstruction of neural tissue from transmission electron microscope (TEM) images has been improved and expanded in functionality and scope. An automated acquisition system captures images of tissue and controls the movement of a TEM. These images comprise a dataset of roughly one gigabyte. Using these data, software running on a Connection Machine automatically reassembles individual images into a single image of each section. An automated contour extraction and object classification algorithm is used and the objects to be reconstructed are selected by the user. Registration is completely automated, but the result is user verifiable and modifiable. The registration parameters are then used to realign both the contour and raw image data. The contour data are smoothed to average out noise, a surface grid is generated, and the resulting reconstruction is visualized. The image data can also be volume visualized. The result is a completely digital, easy-to-use, quantifiable, and generalizable system for 3D reconstruction from transmission electron microscope serial sections.!8