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

Classification of human brain tumours based on texture analysis of label-free multiphoton images

Klassifikation von humanen Hirntumoren basierend auf Texturanalyse markierungsfreier Multiphotonen Bilder

Meeting Abstract

  • presenting/speaker Ortrud Uckermann - Universitätsklinikum Dresden, Neurochirurgie, Dresden, Deutschland
  • Georg Mark - Universitätsklinikum Dresden, Neurochirurgie, Dresden, Deutschland
  • Roberta Galli - TU Dresden, Medizinische Fakultät, Dresden, Deutschland
  • Edmund Koch - TU Dresden, Medizinische Fakultät, Dresden, Deutschland
  • Gabriele Schackert - Universitätsklinikum Dresden, Neurochirurgie, Dresden, Deutschland
  • Gerald Steiner - TU Dresden, Medizinische Fakultät, Dresden, Deutschland
  • Matthias Kirsch - Universitätsklinikum Dresden, 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. DocP155

doi: 10.3205/19dgnc492, urn:nbn:de:0183-19dgnc4929

Published: May 8, 2019

© 2019 Uckermann 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: Multiphoton imaging combining coherent anti-Stokes Raman scattering (CARS) to address lipids, second harmonic generation (SHG) to visualize collagen and two photon excited fluorescence (TPEF) to analyze endogenous fluorophores allows the assessment the tissue’s morphochemistry. Those label-free technologies were suggested for intraoperative tumor recognition and delineation. However, data about interpatient variability and strategies for image analysis are missing so far.

Methods: This study investigated samples of low grade astrocytoma WHO I/II (n=14), anaplastic astrocytoma WHO III (n=73), anaplastic oligodendroglioma WHO III (n=41), GBM (n=91), recurrent GBM (n=18), brain metastases of lung cancer (n=46), of colon cancer (n=25), of renal cancer (n=20), of breast cancer (n=24) and of malignant melanoma (n=29). Multimodal CARS/TPEF/SHG images were obtained from unstained cryosections or fresh biopsies of brain tumors and 29 human non-tumor brains. Texture parameters (mean, standard deviation, kurtosis, skewness, entropy, contrast, correlation, energy, homogeneity) were calculated and linear discriminant analysis was employed for discrimination of tumor versus non-tumor.

Results: Classification resulted in correct rates of 85%, 93% and 57% for CARS, TPEF and SHG respectively. However, the combined analysis of texture parameters of CARS and TPEF images proved to be most suited for the discrimination of non-neoplastic brain versus brain tumors leading to a correct rate of 96% (sensitivity: 96%, specificity: 100%). To approximate the clinical setting, the results were validated on fresh tumor biopsies. 82% (9/11) of the tumors and, most important, all of the non-tumor samples were correctly recognized giving a correct rate of 90% for fresh, unfixed tissue. The majority of single images of each patient’s sample was correctly classified with high probabilities, which is important for clinical translation. Interestingly, an image resolution of 1 µm was sufficient to distinguish brain tumors and non-tumor that will allow fast image acquisition.

Conclusion: Texture analysis of combined CARS/TPEF images of human brain tumors provides reliable information on tissue type even in the context of high intra- and interpatient variability. In combination with miniaturized multiphoton endoscopes, this approach holds great promise for intraoperative brain tumor recognition and delineation and might help the surgeon in optimization of the extent of resection for each patient.