gms | German Medical Science

70. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC)
Joint Meeting mit der Skandinavischen Gesellschaft für Neurochirurgie

Deutsche Gesellschaft für Neurochirurgie (DGNC) e. V.

12.05. - 15.05.2019, Würzburg

Automated segmentation of the subthalamic nucleus – reliable for clinical routine use?

Automatische Segmentierung des Nucleus subthalamicus – geeignet für den klinischen Routinegebrauch?

Meeting Abstract

  • presenting/speaker Amir Zolal - SRH Wald-Klinikum Gera GmbH, Wirbelsäulenchirurgie und Neurotraumatologie, Gera, Deutschland
  • Witold Polanski - Universitätsklinikum Carl Gustav Carus, Klinik und Poliklinik für Neurochirurgie, Dresden, Deutschland
  • Hakan Sitoci - Universitätsklinikum Carl Gustav Carus, Klinik und Poliklinik für Neurochirurgie, Dresden, Deutschland
  • Gabriele Schackert - Universitätsklinikum Carl Gustav Carus, Klinik und Poliklinik für Neurochirurgie, Dresden, Deutschland
  • Stephan B. Sobottka - Universitätsklinikum Carl Gustav Carus, Klinik und Poliklinik für Neurochirurgie, Dresden, Deutschland

Deutsche Gesellschaft für Neurochirurgie. 70. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC), Joint Meeting mit der Skandinavischen Gesellschaft für Neurochirurgie. Würzburg, 12.-15.05.2019. Düsseldorf: German Medical Science GMS Publishing House; 2019. DocP108

doi: 10.3205/19dgnc444, urn:nbn:de:0183-19dgnc4445

Veröffentlicht: 8. Mai 2019

© 2019 Zolal et al.
Dieser Artikel ist ein Open-Access-Artikel und steht unter den Lizenzbedingungen der Creative Commons Attribution 4.0 License (Namensnennung). Lizenz-Angaben siehe http://creativecommons.org/licenses/by/4.0/.


Gliederung

Text

Objective: Various automated segmentation algorithms for the subthalamic nucleus (STN) and other deep brain stimulation targets have been published recently. However, most of the available software is not approved for clinical use. We aimed to validate a clinically available automated segmentation tool of the navigation planning software Brainlab Elements (BL-E) by comparing the output to the gold standard represented by manual segmentation. For comparison, a different method of automated segmentation with recently published DISTAL atlas (DA) and Horn electrophysiological atlas (HEA) was used.

Methods: Preoperative MRI data of 30 Parkinson disease patients were used, resulting in 60 STN segmentations. The segmentations were created manually by two experts. For interrater comparison, 10 STNs were segmented for a second time by the segmenter that did not create the first segmentation. Automated segmentations of STN were obtained from BL-E and ANTs SyN-based atlas segmentation. Dice and Jaccard quotients, target overlap, and false negative/positive values (FNV/FPV) were obtained for the comparison of the manual and automated segmentations. The investigators performing the analysis had no access to the automated STNs at the time of manual segmentation.

Results: For manual segmentation, the mean size of the segmented STN was 133±24 mm3. The mean size of the STN was 121±18 mm3 for Brainlab, 162±21 mm3 for DISTAL atlas and 130±17 mm3 for Horn electrophysiological atlas. The Dice coefficients for the comparison of manual segmentation with BL-E, DA and HEA were 0,54±0,11, 0,59±0,14 and 0,52±0,15 respectively. Using paired Student-t-test with Holm correction, significant differences between BL-E and DISTAL atlases were detected for the Jaccard coefficient (0.38 and 0.43 respectively, p=0.03), target overlap (0.55 and 0.66, p<0.001) and the FNV (0.47 and 0.34, p<0.001). Particularly, there was no significant difference for the Dice coefficient. There were no significant differences between the measures obtained with BL-E and the HEA. The Dice coefficient for the interrater comparison was 0,631.

Conclusion: The automated segmentation algorithm of the newest version of the clinically approved software Brainlab Elements delivers STN segmentations with high precision, with similarity measures only slightly below those for interrater comparison and on par with current high-end not clinically approved approaches.