Shape-based glioma mutation prediction using magnetic resonance imaging

Gliomas are the most frequently occurring primary brain tumors. Determination of the IDH-mutation (Isocitrate DeHydrogenase) in these tumors improves classification and predicts survival. Currently, the only way of determining the mutation status is through a brain biopsy, which is an invasive procedure. This paper concerns the classification of a brain tumor's mutation status through medical imaging. This study proposes a method based on shape description and machine learning. Magnetic resonance images of brain tumors were manually segmented through contour drawing, then analyzed through mathematical shape description. The extracted features were classified using multiple algorithms of which Random Undersampling Boosted Trees gave the highest accuracy. An accuracy of 86.4% was found using leave-one-out cross-validation on a data set of 13 IDH-positive and 9 IDH-wild-type gliomas. The results indicate the feasibility of the proposed approach, but further research on a larger data set is required.