gms | German Medical Science

72. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC)
Joint Meeting mit der Polnischen Gesellschaft für Neurochirurgie

Deutsche Gesellschaft für Neurochirurgie (DGNC) e. V.

06.06. - 09.06.2021

Mapping of spatial heterogeneity of metabolic imaging in glioma using MR-spectroscopy

Mapping der räumlichen Heterogenität der metabolischen Bildgebung bei Gliomen mittels MR-Spektroskopie

Meeting Abstract

  • presenting/speaker Pamela Heiland - Universitätsklinikum Freiburg, Neurochirurgie, Freiburg, Deutschland
  • Irene Hübschle - Albert Ludwigs Universität Freiburg, Medizin, Freiburg, Deutschland
  • Karam Dacca - Albert Ludwigs Universität Freiburg, Medizin, Freiburg, Deutschland
  • Jürgen Beck - Universitätsklinikum Freiburg, Neurochirurgie, Freiburg, Deutschland
  • Horst Urbach - Universitätsklinikum Freiburg, Neuroradiologie, Freiburg, Deutschland
  • Oliver Schnell - Universitätsklinikum Freiburg, Neurochirurgie, Freiburg, Deutschland
  • Irina Mader - Albert Ludwigs Universität Freiburg, Medizin, Freiburg, Deutschland
  • presenting/speaker Dieter Henrik Heiland - Universitätsklinikum Freiburg, Neurochirurgie, Freiburg, Deutschland

Deutsche Gesellschaft für Neurochirurgie. 72. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC), Joint Meeting mit der Polnischen Gesellschaft für Neurochirurgie. sine loco [digital], 06.-09.06.2021. Düsseldorf: German Medical Science GMS Publishing House; 2021. DocV148

doi: 10.3205/21dgnc143, urn:nbn:de:0183-21dgnc1432

Veröffentlicht: 4. Juni 2021

© 2021 Heiland 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: In neuro-oncology, the necessity to examine the biological and molecular properties of glial brain tumors for accurate diagnosis has become crucial. In our prospective study, we aimed to investigate the predictive value of MR-spectroscopy in order to establish a solid automated preoperative molecular characterization of these tumors. We developed a novel MR-spectroscopy-based algorithm that allows the specific molecular analysis of these tumors through a radiomics analytics pipeline.

Methods: 90 patients with assumed high- and low-grade brain lesions were enrolled in our prospective imaging trial from 2016 to 2019 and received preoperative anatomical and multi-voxel MRS (5x5x15mm3 voxel size). Tumor regions were segmented and co-registered to corresponding spectroscopic voxels. Investigation of spatial diversity in tumor-associated metabolic architecture was performed by high-dimensional computational approaches implemented in an innovative R-based package of MRS data analysis. For prediction of molecular profiles based on MRS imaging, we used the bottleneck layer of a deep autoencoder in a multi-layer linear discriminant analysis to predict the molecular profile based on MRS imaging.

Results: Cluster analysis of all spectra revealed 10 different clusters, 6 clusters contained predominantly normal appearing matter, 4 clusters contained FLAIR hyperintense regions, 2 of which were defined as tumor-associated. In the FLAIR hyperintense regions, tumor-infiltrative regions marked by altered choline intensity were separated from edema with a defined signal depression due to increased water peeks. A deep learning model showed high accuracy in differentiating tumor and non-malignant spectra (97.3%). Prediction of the molecular subtype of glioma was 83.4% with the weakest performance in oligodendroglioma (76.7%). Considering the prediction within a spatial resolved MRS dataset, we combined the single-voxel prediction with an MRS-based classifier score which revealed an overall accuracy of 91.2%.

Conclusion: Our data suggest that MRS imaging is a powerful tool to distinguish infiltrative and edema regions and predict non-invasively the molecular profile of brain tumours. The second part of our prospective imaging trial will focus on the validation of our MRS-based classifier.