Automated Reconstruction of Dendritic and Axonal Trees by Global Optimization with Geometric Priors
click on the images to jump to some results.
Tree-like structures, such as dendritic, vascular, or bronchial networks, are pervasive in biological systems.
With the advent of modern acquisition techniques that produce endless streams of imagery, there has been renewed
interest in automated delineation to exploit this data. However, despite many years of sustained effort, automated
techniques remain fragile and error-prone. In this work, we use 3D optical micrographs of neurons and 2D retinal fundus
images to demonstrate the importance of taking global tree structure and geometry into account to improve topological
accuracy of the delineations.
Unlike earlier methods that rely mostly on local evidence, ours builds a set of candidate trees over many different subsets
of points likely to belong to the optimal tree and then chooses the best one according to a global objective function that
combines image evidence with geometric priors. Since the best tree does not necessarily span all the points, the algorithm
is able to eliminate false detections while retaining the correct tree topology.
This animation walks through the intermediate steps
of our algorithm on both Olfactory Projection Fibers of the
DIADEM
challenge and the retinal scans of the
DRIVE database. Select a data
type from the left menu and click on the images to advance the
slideshow.
Results
Reconstructions for the Neocortical Layer 1 Axons of the DIADEM Challenge [top]
This stack is obtained by labeling axons from Neocortical Layer 1 of mouse brain using GFP and imaging them
with a two-photon microscope. Each colored tree in the video corresponds to different root node.
For additional details on the dataset, please refer to the DIADEM website.
A reconstruction of the Olfactory Projection Fibers of the DIADEM Challenge [top]
This stack is acquired from the olfactory bulb of Drosophila fly using 2-channel confocal microscopy.
Our reconstruction is overlaid in red. The yellow sphere is the tree root, which is manually selected.
For additional details on the dataset, please refer to the DIADEM website.
This image stack is acquired by our colleagues from EPFL's Brain Mind Institute.
The images were obtained from biocityne-dyed rat brains. The numerous artifacts produced by
irregularities of the staining process and the non-Gaussian blur introduced by the microscope
make their automated analysis challenging. Furthermore, many significant processes appear as
faint structures, present abrupt intensity changes, or are severely blurred. Our algorithm
eliminates both structured and unstructured noise, and produces a topologically plausible tree.
This animation provides a visual comparison of our reconstructions, manually annotated
ground truths and Minimum Spanning Tree reconstructions for a retinal scan and a brigthfield
image stack. Select a data type from the left menu and click on the images to
advance the slideshow.
We will be sharing our software here in future. Stay tuned.
References
E. Turetken, G. Gonzalez, C. Blum and P. Fua, Automated Reconstruction of Dendritic and Axonal Trees by Global Optimization with Geometric Priors, Neuroinformatics, In Press, 2011.
