Deep Learning Segmentation of Optical Microscopy Images Improves 3-D Neuron Reconstruction

Abstract

Digital reconstruction, or tracing, of 3-D neuron structure from microscopy images is a critical step toward reversing engineering the wiring and anatomy of a brain. Despite a number of prior attempts, this task remains very challenging, especially when images are contaminated by noises or have discontinued segments of neurite patterns. An approach for addressing such problems is to identify the locations of neuronal voxels using image segmentation methods, prior to applying tracing or reconstruction techniques. This preprocessing step is expected to remove noises in the data, thereby leading to improved reconstruction results. In this paper, we proposed to use 3-D convolutional neural networks (CNNs) for segmenting the neuronal microscopy images. Specifically, we designed a novel CNN architecture, that takes volumetric images as the inputs and their voxel-wise segmentation maps as the outputs. The developed architecture allows us to train and predict using large microscopy images in an end-to-end manner. We evaluated the performance of our model on a variety of challenging 3-D microscopy images from different organisms. Results showed that the proposed methods improved the tracing performance significantly when combined with different reconstruction algorithms.

Identifying weak signals in inhomogeneous neuronal images for large-scale tracing of neurites

“…ST-LFV involves more rules deduced from images by human beings than other methods (R. Li et al 2017;Chen et al 2015). These rules are based on the fact that our premises are commonly applicable to most neuronal images collected with optical microscopes, and provide a basis for constructing a feature vector that displays the differences between foreground and background voxels.…”

“…In ST-LFV, the identification model is embedded into the tracing procedure, which is different from other methods (R. Li et al 2017;Chen et al 2015). When the tracing termination conditions are effective, the identification method works and identifies whether the current tracing point is a foreground voxel.…”

“…However, most of these methods use a set of thresholds to determine whether the traced neurite continues or not, and thus may encounter difficulty separating weak signals from an inhomogeneous background. Recently proposed machine learning methods (R. Li et al 2017;Chen et al 2015;Megjhani et al 2015;Gu et al 2017;Hernandez-Herrera et al 2014;Becker et al 2013) can generate tracing results with better accuracy than traditional methods. However, these methods may fail to rapidly trace neurites in large-scale images.…”

“…Machine learning methods consider many image features and use an algorithm to detect dominant image features, thereby requiring intensive computations. Some of these methods instead separate the identification and tracing procedures (R. Li et al 2017;Chen et al 2015;Megjhani et al 2015) and attempt to identify as many signal voxels as possible, which generates the detailed shape of a neurite. However, heavier computation costs are incurred for larger images.…”

DeepNeuron: An Open Deep Learning Toolbox for Neuron Tracing

“…Chen et al proposed a self-learning based tracing approach, which did not require substantial human annotations (Chen, et al, 2015). Fakhry et al (Fakhry, et al, 2016) and Li et al (Li, et al, 2017) used deep learning neural networks to segment electron and light microscopy neuron images. Despite these algorithmic efforts, none of these methods provide publicly available tools to use on external datasets.…”