Automation of 3D reconstruction of neural tissue from large volume of conventional serial section transmission electron micrographs

Abstract

We describe an approach for automation of the process of reconstruction of neural tissue from serial section transmission electron micrographs. Such reconstructions require 3D segmentation of individual neuronal processes (axons and dendrites) performed in densely packed neuropil. We first detect neuronal cell profiles in each image in a stack of serial micrographs with multi-scale ridge detector. Short breaks in detected boundaries are interpolated using anisotropic contour completion formulated in fuzzy-logic framework. Detected profiles from adjacent sections are linked together based on cues such as shape similarity and image texture. Thus obtained 3D segmentation is validated by human operators in computer-guided proofreading process. Our approach makes possible reconstructions of neural tissue at final rate of about 5µm 3 /man-hour, as determined primarily by the speed of proofreading. To date we have applied this approach to reconstruct few blocks of neural tissue from different regions of rat brain totaling over 1000µm 3 , and used these to evaluate reconstruction speed, quality, error rates, and presence of ambiguous locations in neuropil ssTEM imaging data.

Segmentation fusion for connectomics

“…A growing number of efforts have been dedicated recently to the problem of 3D labeling individual neurons from full stacks [8,9,20,14]. Some of these solutions are based on Markov Random Fields (MRFs) at the pixel level.…”

“…A common approach in addressing this problem is to first obtain a 2D segmentation of each section, and then to match or agglomeratively cluster these 2D segments across sections in the stack [9,20,14]. However, such methods rely on the assumption that the initial 2D segmentation of each section is good enough for the posterior grouping, or that every neuron has been oversegmented [20,9].…”

“…However, such methods rely on the assumption that the initial 2D segmentation of each section is good enough for the posterior grouping, or that every neuron has been oversegmented [20,9]. In addition, some of these methods need indirect penalties to prevent trivial yet incorrect clusterings [20], or require setting stopping conditions or manually-designed rules to converge to the right solution [9,14]. Finally, a number of methods rely on greedy segment-merging strategies [9,14] and cannot guarantee global clustering optimality.…”

A Bayesian compressed-sensing approach for reconstructing neural connectivity from subsampled anatomical data

“…However, data acquisition in this setting remains difficult, as is the analysis of the resulting very large imaging data sets (Burton, 1999;Briggman and Denk, 2006;Smith, 2007;Helmstaedter et al, 2008;Lu et al, 2009;Helmstaedter et al, 2011). The automation of data acquisition, image analysis, and neural process tracking in serial electron microscopy remains an area of active research (Denk and Horstmann, 2004;Smith, 2007;Helmstaedter et al, 2008;Mishchenko, 2009;Lu et al, 2009;Briggman et al, 2011;Bock et al, 2011;Helmstaedter et al, 2011).…”

An Interactive Editing Framework for Electron Microscopy Image Segmentation

“…Therefore, researchers have put in a great amount of effort in developing segmentation algorithms to extract neuronal morphologies from stacks of serial EM images [1,2,3,4,5,6]. A great majority of these existing segmentation algorithms are designed for the automation of the segmentation pipeline.…”