Assigning a new glioma grade label ground-truth for the BraTS dataset using radiologic criteria

Extensive development of both non-invasive multiparametric magnetic resonance examinations and machine learning tools is challenging the World Health Organization (WHO) pathologic classification (2016). The classification of glioma gradings needs to discriminate between Low Grade Gliomas (LGG), grades I, II; and High Grade Gliomas (HGG), grades III, IV, leading to major issues in oncology for therapeutic management of patients. A well-known dataset for machine-based grade prediction is the MICCAI Brain Tumor Segmentation (BraTS) dataset. However this dataset is not divided into WHO-defined LGG and HGG, since it combines grades I, II and III as "lower grades gliomas", while its HGG category only presents grade IV glioblastoma multiform. In this paper we first investigate the consistency of the original BraTS labels with radiologic criteria using machine-aided predictions. Prediction scores show that our classifier is able to identify the radiologically high grade patients among the original lower grade population of the dataset. This highlights the coherence of radiologic criteria for low grade versus high grade classification under WHO terms. Then we asked 5 expert radiologists to annotate BraTS images between low (as opposed to lower) grade and high grade glioma classes, resulting in a new ground-truth. These labels are based upon radiologic clues such as necrosis or gadolinium enhancement patterns. We then study the feasibility of machine-aided classification into LGG and HGG from purely radiologic features.