gms | German Medical Science

73. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC)
Joint Meeting mit der Griechischen Gesellschaft für Neurochirurgie

Deutsche Gesellschaft für Neurochirurgie (DGNC) e. V.

29.05. - 01.06.2022, Köln

Automated identification of tumour infiltrated brain tissue using optical coherence tomography and deep learning

Automatisierte Erkennung von tumorinfiltriertem Gehirngewebe mittels optischer Kohärenztomographie und deep learning

Meeting Abstract

  • presenting/speaker Paul Strenge - Medizinisches Laserzentrum Lübeck GmbH, Lübeck, Deutschland
  • Birgit Lange - Medizinisches Laserzentrum Lübeck GmbH, Lübeck, Deutschland
  • Christin Grill - Universität zu Lübeck, Institut für Biomedizinische Optik, Lübeck, Deutschland
  • Wolfgang Draxinger - Universität zu Lübeck, Institut für Biomedizinische Optik, Lübeck, Deutschland
  • Sonja Spahr-Hess - Universitätsklinikum Schleswig-Holstein, Neurochirugie Campus Lübeck, Lübeck, Deutschland
  • Patrick Kuppler - Universitätsklinikum Schleswig-Holstein, Neurochirugie Campus Lübeck, Lübeck, Deutschland
  • Veit Danicke - Medizinisches Laserzentrum Lübeck GmbH, Lübeck, Deutschland
  • Dirk Theisen-Kunde - Medizinisches Laserzentrum Lübeck GmbH, Lübeck, Deutschland
  • Heinz Handels - Universität zu Lübeck, Institut für Medizinische Informatik, Lübeck, Deutschland
  • Mario Matteo Bonsanto - Universitätsklinikum Schleswig-Holstein, Neurochirugie Campus Lübeck, Lübeck, Deutschland
  • Christian Hagel - Universitätsklinikum Hamburg-Eppendorf, Institut für Neuropathologie, Hamburg, Deutschland
  • Robert Huber - Universität zu Lübeck, Institut für Biomedizinische Optik, Lübeck, Deutschland
  • Ralf Brinkmann - Medizinisches Laserzentrum Lübeck GmbH, Lübeck, Deutschland

Deutsche Gesellschaft für Neurochirurgie. 73. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC), Joint Meeting mit der Griechischen Gesellschaft für Neurochirurgie. Köln, 29.05.-01.06.2022. Düsseldorf: German Medical Science GMS Publishing House; 2022. DocV083

doi: 10.3205/22dgnc086, urn:nbn:de:0183-22dgnc0863

Published: May 25, 2022

© 2022 Strenge et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License. See license information at http://creativecommons.org/licenses/by/4.0/.


Outline

Text

Objective: The identification of tumor infiltrated brain tissue during neurosurgical resection is still a major challenge in the field of neurooncology. The challenge is aggravated by the infiltrative growth of certain tumor types, like glioblastoma multiforme, which leads to ill-defined tumor borders. Optical coherence tomography (OCT) has proven to be a possible additional imaging method for the identification of tumor and healthy brain tissue. The aim of the project was the identification of different grades of tumor infiltrated brain tissue based on ex vivo OCT B-scans. For the classification, a deep neural network was configured in order to discriminate the different stages of tumor infiltration from healthy tissue.

Methods: Over the course of a clinical study around 160 samples were excised from 17 patients. Samples were extracted from the brain surface, the main tumor mass and from the border of the resection cavity after the resection was finished by the surgeon. Each sample was imaged ex vivo by an OCT system with an imaging wavelength of 1300 nm and was assessed by a neuropathologist. Each histological sections was labelled with either healthy white matter or one of three different tumor infiltration labels (>0-30%, 30-60%, >60% tumor infiltration). For each histological section corresponding OCT B-scans were determined and the labels were transferred onto the OCT B-scan. This resulted in 650 labelled OCT B-scans, which were used as a ground-truth dataset for the supervised classification.

A deep neural network was used for the classification (Figure 1 [Fig. 1]). The network used averaged A-scans from each OCT B-scan as an input, which allowed the network to extract structural and optical properties from the input. For the evaluation a k-fold cross validation was used, where each patient was once used as the test data, while the remaining patients were used for the training.

Results: Table 1 [Tab. 1] shows the test results for overall test data for the classification of different stages of tumor infiltration. It is shown, that the network achieved good classification results. The performance of the network reduces with decreasing grade of tumor infiltration.

Conclusion: OCT is able to discriminate different stages of tumor infiltration from healthy white matter tissue with reasonable accuracy. The used OCT system was transferred into a surgical microscope in order to apply the neural network on in vivo data in the future.