gms | German Medical Science

61. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC) im Rahmen der Neurowoche 2010
Joint Meeting mit der Brasilianischen Gesellschaft für Neurochirurgie am 20. September 2010

Deutsche Gesellschaft für Neurochirurgie (DGNC) e. V.

21. - 25.09.2010, Mannheim

Evaluation of several cost functions for Min-Cut-segmentation of fiber bundles in the human brain

Meeting Abstract

  • Miriam H. Bauer - Klinik für Neurochirurgie, Universitätsklinikum Marburg, Deutschland; FB Mathematik & Informatik, Universität Marburg, Deutschland
  • Jan Egger - Klinik für Neurochirurgie, Universitätsklinikum Marburg, Deutschland; FB Mathematik & Informatik, Universität Marburg, Deutschland
  • Daniela Kuhnt - Klinik für Neurochirurgie, Universitätsklinikum Marburg, Deutschland
  • Sebastiano Barbieri - Fraunhofer MEVIS, Bremen, Deutschland
  • Bernd Freisleben - FB Mathematik & Informatik, Universität Marburg, Deutschland
  • Christopher Nimsky - Klinik für Neurochirurgie, Universitätsklinikum Marburg, Deutschland

Deutsche Gesellschaft für Neurochirurgie. 61. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC) im Rahmen der Neurowoche 2010. Mannheim, 21.-25.09.2010. Düsseldorf: German Medical Science GMS Publishing House; 2010. DocP1821

doi: 10.3205/10dgnc292, urn:nbn:de:0183-10dgnc2921

Veröffentlicht: 16. September 2010

© 2010 Bauer et al.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen (http://creativecommons.org/licenses/by-nc-nd/3.0/deed.de). Er darf vervielfältigt, verbreitet und öffentlich zugänglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.


Gliederung

Text

Objective: Intraoperative imaging and multimodal navigation improved glioma surgery in the last decades. Functional data, integrated in the neuronavigation procedure, has contributed to safe tumor resection for lesions located near eloquent brain structures. Functional MRI (fMRI) and magneto encephalography (MEG) have been integrated into the navigation to identify eloquent cortical brain areas. In addition, visualising white matter tracts is now possible by diffusion tensor imaging (DTI) segmentation. To optimize the displayed objects, a new graph-based segmentation method for fiber bundles and different cost functions for graph construction are presented in this paper.

Methods: Based on a centerline of the fiber bundle derived form fiber tracking, rays are sent out and sampled in different planes perpendicular to the centerline. Then, a directed and weighted graph is created containing all evaluation points and a sink and a source node. Besides infinit-weighted edges connecting the nodes, different cost functions are used to weigh the edges to source and sink. Finally, a minimum cost closed set is computed via a polynomial time s-t-cut, which creates an optimal segmentation of the fiber bundle. This result is transformed into a 3D object by triangulation and voxelization. As cost functions, the scalar measures fractional anisotropy (FA), mean diffusivity (MD) and relative anisotropy (RA) are used. The algorithm has been implemented in C++ within the MeVisLab platform on an Intel Core2Quad CPU, 3 GHz, 6 GB RAM.

Results: The evaluation is based on DTI data sets of several software-phantoms (portion of torus with different noise levels, anatomical phantom with modelled corticospinal tract) to have ground truth data (fiber location, course) to compare against. As our quality measure, the Dice Similarity Coefficient (DSC) is calculated. Using the same parameter combination for all determinations, the cost function using FA yielded an average DSC of 73.58±0,047%, while the average DSC of using MD was 79.16±0.037% and for RA it was 77,05±0.015%.

Conclusions: A new graph-based min-cut-segmentation for fiber bundle segmentation was evaluated by using different cost functions for graph construction. In conclusion, exact and automatic segmentation of the fiber bundles in the human brain obtained by our novel approach is useful in glioma resection, leading to safe tumor resection for lesions located near eloquent brain structures.