Support Vector Machines in DSC-Based Glioma Imaging Suggestions for Optimal Characterization

Frank G. Zllner¹, Kyrre Eeg Emblem², Lothar R. Schad¹

Dynamic susceptibility contrast magnetic resonance perfusion imaging (DSC-MRI) is a method of choice to characterize gliomas. Recently, support vector machines (SVM) have been introduced as means to prospectively characterize new patients based on information from previous patients. Based on features derived from automatically segmented tumor volumes from 101 DSC-MR examinations, four different SVM models were compared. All SVM models achieved high prediction accuracies (>82%) after rebalancing the training data sets to equal amounts of samples per class. Best discrimination were obtained using a SVM model with a radial basis function kernel allowing for a correct prediction of low-grade glioma at 83% and high-grade glioma at 91%.