Benjamn Garzn1, Kyrre E. Emblem2,3, Kim Mouridsen4, Atle Bjrnerud2,5, Asta Hberg6, Yngve Kvinnsland7
1Department of Circulation and Medical Imaging, NTNU, Trondheim, Norway; 2Department of Medical Physics, Rikshospitalet University Hospital, Oslo, Norway; 3The Interventional Centre, Rikshospitalet University Hospital , Oslo, Norway; 4Center of Functionally Integrative Neuroscience, Aarhus University, Aarhus, Denmark; 5Department of Physics, University of Oslo , Oslo, Norway; 6Department of Circulation and Medical Imaging, NTNU, Trondheim, Norway; 7Department of Surgical Sciences, University of Bergen, Bergen, Norway
The purpose of the study was to investigate multiple MR-derived image features with respect to diagnostic accuracy in glioma grading. Structural and dynamic susceptibility contrast scans of 74 glioma patients (29 low grade, 45 high grade) with histologically confirmed grade were assessed. Perfusion maps were derived and a set of features was calculated at ROIs encompassing the tumor. For each combination of up to 5 features, a logistic regression model was fitted to a balanced and random training set. Model performance was assessed by applying the model to the remaining subjects and calculating the median area under the ROC curve with a bootstrap procedure. The combination yielding the best performance was maximum enhancement, roundness and skewness of time-to-peak of the first-pass perfusion curve.