Statistics-Driven Localization of Dissimilarities in Data

Information

• Publication Type: Technical Report
• Workgroup(s)/Project(s):
  • Visualization Group
• Date: April 2016
• Number: TR-186-2-16-1

Abstract

The identification of dissimilar regions in spatial and temporal data is a fundamental part of data exploration. This process takes place in applications, such as biomedical image processing as well as climatic data analysis. We propose a general solution for this task by employing well-founded statistical tools. From a large set of candidate regions, we derive an empirical distribution of the data and perform statistical hypothesis testing to obtain p-values as measures of dissimilarity. Having p-values, we quantify differences and rank regions on a global scale according to their dissimilarity to user-specified exemplar regions. We demonstrate our approach and its generality with two application scenarios, namely interactive exploration of climatic data and segmentation editing in the medical domain. In both cases our data exploration protocol unifies the interactive data analysis, guiding the user towards regions with the most relevant dissimilarity characteristics. The dissimilarity analysis results are conveyed with a radial tree, which prevents the user from searching exhaustively through all the data.

Additional Files and Images

Weblinks

No further information available.

BibTeX

@techreport{karimov-2016-SD,
  title =      "Statistics-Driven Localization of Dissimilarities in Data",
  author =     "Alexey Karimov and Gabriel Mistelbauer and Thomas Auzinger
               and Eduard Gr\"{o}ller",
  year =       "2016",
  abstract =   "The identification of dissimilar regions in spatial and
               temporal data is a fundamental part of data exploration.
               This process takes place in applications, such as biomedical
               image processing as well as climatic data analysis. We
               propose a general solution for this task by employing
               well-founded statistical tools. From a large set of
               candidate regions, we derive an empirical distribution of
               the data and perform statistical hypothesis testing to
               obtain p-values as measures of dissimilarity. Having
               p-values, we quantify differences and rank regions on a
               global scale according to their dissimilarity to
               user-specified exemplar regions. We demonstrate our approach
               and its generality with two application scenarios, namely
               interactive exploration of climatic data and segmentation
               editing in the medical domain. In both cases our data
               exploration protocol unifies the interactive data analysis,
               guiding the user towards regions with the most relevant
               dissimilarity characteristics. The dissimilarity analysis
               results are conveyed with a radial tree, which prevents the
               user from searching exhaustively through all the data.",
  month =      apr,
  number =     "TR-186-2-16-1",
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  institution = "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology ",
  note =       "human contact: technical-report@cg.tuwien.ac.at",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2016/karimov-2016-SD/",
}