Visual computing techniques for the reconstruction and analysis of anatomically realistic neural networks

  • To understand how the brain translates sensory input into behavior, one needs to identify, at the cellular level, the involved neural circuitry and the electrical signals it carries. This thesis describes methods and tools that enable neuroscientists to obtain important anatomical data, including neuron numbers and shapes, from 3D microscopy images. On this basis, tools have been developed to create and visually analyze anatomically realistic 3D models of neural networks: 1. An automatic segmentation method for determining the number and location of neuron cell bodies in 3D microscopy images. Application of this method yields a difference of merely ∼4% between automatically and manually counted cells, which is sufficiently accurate for application in large-scale counting experiments. 2. A method for the automatic alignment of 3D section volumes containing filamentous structures. To this end, an existing point-matching-based method has been adapted such that sections containing neuron and microtubule fragments could be successfully aligned. 3. The Filament Editor, a 3D proof-editing tool for visual verification and correction of automatically traced filaments. The usefulness of the Filament Editor is demonstrated by applying it in a validated neuron reconstruction pipeline to create 3D models of long-range and complex neuronal branches. 4. The tool NeuroNet, which is used to assemble an anatomical model of a neural network representing the rat barrel cortex (or subnetworks therein, e.g. individual cortical columns), based on reconstructed anatomical data, such as neuron distributions and 3D morphologies. The tool estimates synaptic connectivity between neurons based on structural overlap between axons and dendrites. 5. A framework for the interactive visual analysis of synaptic connectivity in such networks at multiple scales. It works from the level of neuron populations down to individual synapse positions on dendritic trees. It comprises the Cortical Column Connectivity Viewer, developed to analyze synaptic connections between neuron populations within and between cortical columns. The usefulness of these methods is demonstrated by applying them to reconstruct and analyze neural networks in the rat barrel cortex. Finally, I describe several applications of these methods and tools by neuroscientists, yielding significant biological findings regarding neuron anatomy and connectivity.
Metadaten
Author:Vincent J. Dercksen
Document Type:Doctoral Thesis
Tag:3D reconstruction; connectome; image segmentation; neural network; neuron tracing
CCS-Classification:E. Data
Granting Institution:Freie Universität Berlin
Advisor:Christof Schütte, Markus Hadwiger
Date of final exam:2015/11/12
Date of first Publication:2015/12/11
URL:http://www.diss.fu-berlin.de/diss/receive/FUDISS_thesis_000000100938