COMP-12. TOWARDS BIG DATA IN DIGITAL NEUROPATHOLOGY WITH THE DIGITAL BRAIN TUMOR ATLAS. (5th November 2018)

• Record Type: Journal Article
• Authors: Roetzer, Thomas; Moser, Anna-Christina; Mercea, Petra; Prihoda, Romana; Atli, Baran; Hainfellner, Johannes; Baumann, Bernhard; Langs, Georg; Woehrer, Adelheid
• Is Part Of: Neuro-oncology. Volume 20(2018)Supplement 6
• Page Start: vi66
• Page End: vi66
• Publication Date: 2018-11-05
• Subjects: Brain Neoplasms -- Periodicals; Brain -- Tumors -- Periodicals; Brain -- Cancer -- Periodicals; Nervous system -- Cancer -- Periodicals; 616.99481
• DOI: 10.1093/neuonc/noy148.267
• Languages: English
• ISSNs: 1522-8517

Abstract

Abstract: Digital methods are increasingly applied to enhance disease classification, tissue segmentation, and prognostic profiling in virtually all medical specialties. They are recognized also for informing precision diagnostics in neuropathology. However, most computational algorithms depend on large training data sets, which are not yet readily available for tissue slides. Regarding brain tumors, only single histological repositories are freely available, which cover only few tumor types and are limited by poor tissue morphology. Thus, in order to fill this gap, we are in the process of establishing the 'Digital Brain Tumor Atlas', which will span the full diagnostic spectrum of brain tumors according to the latest WHO classification. The repository will be based on digitized slides from MedUni Vienna's neurobiobank, which is one of the largest in Europe containing well over 500, 000 surgical samples from 1882 to 2018. We are currently scanning a total of 10, 000 diagnostic H&E-stained brain tumor slides at high resolution using a Hamamatsu NanoZoomer XR scanner. To date, roughly 2, 700 whole slide images that cover 102 different brain tumor types have already been digitized. All slides will be annotated with tumor type (including molecular subtype where applicable) and location as well as patient age and sex. We will put special focus on the group of diffuse gliomas, for which we have already scanned 500 slides, each providing expert segmentations for compact tumor, necrosis, hemorrhage, and pre-existing brain tissue. In the future, this unique brain tumor slide repository will be made freely available online to enhance further research in digital neuropathology by 1, providing all data necessary for training meaningful machine learning algorithms and 2, comprising an independent dataset for external validation purposes.