Noise and Artifact Reduction in Interactive Volume Renderings of Electron-Microscopy Data-Sets

Information

• Publication Type: Master Thesis
• Workgroup(s)/Project(s):
  • Visualization Group
• Date: May 2010
• Diploma Examination: 15. May 2010
• First Supervisor:
  • Eduard Gröller
  • Markus Hadwiger
  • Johanna Beyer

Abstract

Connectomics is an emerging area of neuroscience that is concerned with understanding the neural algorithms embeded in the neural circuits of the brain by tracking neurons and studying their connections. From all the available scanning technologies only electron microscopy (EM) can provide sufficient scanning resolutions in order to identify neural processes. EM data sets, however, suffer from bad signal-to-noise ratio and artifacts introduced to the data set during the sectioning and digital reconstruction process of the scanned specimen. In this thesis we present two different approaches that generally allow noise and artifact reduction on volumetric data sets and which can be used to increase the visual quality of direct volume renderings (DVRs) of EM data sets. The fist approach we developed was an interactive, on-the-fly filtering framework that allows a user to filter even very large volume data set with resizable 3D filter-kernels. For comparison, we implemented an average, a Gaussian, and a bilateral filter. The second approach we investigated is a semi-automatic one that allows a user to select regions within a data set. Similar regions are then retrieved by our algorithm using multiresolution histograms and the user can remove these regions from the rendering. By selecting and hiding regions containing noise or artifacts, the desired noise- and artifact-reduction can be achieved. We are going to show that both methods we investigated are suitable for removing noise and artifacts in EM data sets.

Additional Files and Images

Weblinks

No further information available.

BibTeX

@mastersthesis{ritzberger-2010-nar,
  title =      "Noise and Artifact Reduction in Interactive Volume
               Renderings of Electron-Microscopy Data-Sets",
  author =     "Andreas Ritzberger",
  year =       "2010",
  abstract =   "Connectomics is an emerging area of neuroscience that is
               concerned with understanding the neural algorithms embeded
               in the neural circuits of the brain by tracking neurons and
               studying their connections. From all the available scanning
               technologies only electron microscopy (EM) can provide
               sufficient scanning resolutions in order to identify neural
               processes. EM data sets, however, suffer from bad
               signal-to-noise ratio and artifacts introduced to the data
               set during the sectioning and digital reconstruction process
               of the scanned specimen. In this thesis we present two
               different approaches that generally allow noise and artifact
               reduction on volumetric data sets and which can be used to
               increase the visual quality of direct volume renderings
               (DVRs) of EM data sets. The fist approach we developed was
               an interactive, on-the-fly filtering framework that allows a
               user to filter even very large volume data set with
               resizable 3D filter-kernels. For comparison, we implemented
               an average, a Gaussian, and a bilateral filter. The second
               approach we investigated is a semi-automatic one that allows
               a user to select regions within a data set. Similar regions
               are then retrieved by our algorithm using multiresolution
               histograms and the user can remove these regions from the
               rendering. By selecting and hiding regions containing noise
               or artifacts, the desired noise- and artifact-reduction can
               be achieved. We are going to show that both methods we
               investigated are suitable for removing noise and artifacts
               in EM data sets.",
  month =      may,
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology ",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2010/ritzberger-2010-nar/",
}