Skip to main navigation Skip to search Skip to main content

Automatic Vertebrae Segmentation in MR Volumes A Comparison of Different Deep Learning-based Approaches

  • Orgest Xhelili
  • , Miruna Gafencu
  • , Francesca De Benetti
  • , Nassir Navab
  • , Thomas Wendler
  • Technical University of Munich
  • SurgicEye GmbH

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

2 Scopus citations

Abstract

Vertebrae segmentation is important in several clinical settings involving spine pathologies. In recent years, deep learning-based techniques have become popular in segmenting the spine in computed tomography (CT) volumes. However, few options have been tested for magnetic resonance (MR) imaging segmentation. In this paper, we provide a comparison of three deep learning methods tackling the automatic vertebrae segmentation in MR volumes. We selected three methods that were already established in the segmentation of CT images: 3D UNet as our baseline, an iterative binary segmentation approach, and a multi-stage segmentation approach. Our experiments achieved a mean Dice score of 88.1% and demonstrate that CT segmentation methods are easily transferable to MR segmentation.

Original languageEnglish
Title of host publicationBildverarbeitung für die Medizin 2023 Proceedings, German Workshop on Medical Image Computing, Braunschweig
EditorsThomas M. Deserno, Heinz Handels, Andreas Maier, Klaus Maier-Hein, Christoph Palm, Thomas Tolxdorff
PublisherSpringer Science and Business Media Deutschland GmbH
Pages19-24
Number of pages6
ISBN (Print)9783658416560
DOIs
StatePublished - 2023
EventBildverarbeitung für die Medizin Workshop, BVM 2023 - Braunschweig, Germany
Duration: 2 Jul 20234 Jul 2023

Publication series

NameInformatik aktuell
ISSN (Print)1431-472X

Conference

ConferenceBildverarbeitung für die Medizin Workshop, BVM 2023
Country/TerritoryGermany
CityBraunschweig
Period2/07/234/07/23

Fingerprint

Dive into the research topics of 'Automatic Vertebrae Segmentation in MR Volumes A Comparison of Different Deep Learning-based Approaches'. Together they form a unique fingerprint.

Cite this