Kyrre E. Emblem1,2, Frank G. Zoellner3, Atle Bjornerud1,4
1Department of Medical Physics, Rikshospitalet University Hospital, Oslo, Norway; 2The Interventional Centre, Rikshospitalet University Hospital, Oslo, Norway; 3Department of Assisted Clinical Medicine, University of Heidelberg, Mannheim, Germany; 4Department of Physics, University of Oslo, Oslo, Norway
We have assessed whether a fully automated, multi-parametric model for predicting outcome in glioma patients from dynamic susceptibility contrast MR imaging can be used as a second reference to pathologic findings. Based on automatically segmented tumor regions, 3D scatter diagrams of cerebral blood volume as a function of Ktrans were derived for each patient. A predictive model based on support vector machines was used to predict outcome in each patient using scatter diagrams and survival status of the remaining patients. Our results suggest that the proposed approach provides similar diagnostic accuracy values to histopathology when predicting patient outcome.