Segmentation and Comparison of Drosophila Neurons

As we step into an age of emerging research in Bioimage Informatics, a relevent problem in this field is that of analyzing the neuronal structure of complex organisms. While the neurome of the C.elegans have been fully understood, a major research focus is to analyze the neural network of more complicated species. A major challenge in this field is to analyse 3-D microscopy images to represent the neuronal morphology mathematically. Our research in this field focuses on obtaining the neuronal tree from a 3-D confocal microscopy image. The automated neuron tracers Tree2Tree and Tree2Tree2 have been developed to aid neural tracing.


Automated segmentation of neurons is one of the critical open problems in neurobiological image analysis. In order to validate hypotheses regarding the neural connectivity and function of the brain, there exists an ongoing effort to build atlases and libraries of neural structures. The major stages in this effort are to define the morphology of the neurons, the neurome, and the connectivity, the connectome. Such atlases would yield both shape and structure that aids in understanding how cellular structure regulates brain function.

Currently, most of the commercially available neuron tracing softwares are semi-automated, where a trained user is required to manually choose seed-points that belongs to the neuronal portion. Our aim is to automate this tedious process. The low contrast and low SNR of the confocal microscopy images make this task a challenging one. We model this segmentation problem in a graph theoretic framework, where the neuronal tree is visualized as a graph theoretic tree. This abstraction preserves the morphology in the sense that the parent-child connectivity of the neuron voxels is preserved. A Hessian based analysis of the original image is used to enhance the filamentous neuronal portions, which is then binarized by using a local thresholding technique based on a surface growing approach. The connectivity between the disjoint components is established by a novel global connectivityalgorithm. Our method also allows the user to prune some unwanted edges that may have been present due to the inherent clutter in the microscopy images. The neuronal morphology is embedded in a '.swc' file format, which can be easily visualized by using commonly used visualization software.

The following figure shows some tracing results:
neuron compare.JPG
green: Tracing results, violet: ground truth

Research Groups

Topic revision: r1 - 02 Jun 2014, AndreaVaccari
©2017 University of Virginia. Privacy Statement
Virginia Image and Video Analysis - School of Engineering and Applied Science - University of Virginia
P.O. Box 400743 - Charlottesville, VA - 22904 - E-Mail