Prediction of glioblastoma multiforme response to bevacizumab treatment using diffusion and perfusion imaging

Abstract

Response to Bevacizumab in Glioblastoma Multiforme (GBM) is not the same for all patients who receive this antiangiogenic therapy. In this research, structural magnetic resonance images (MRI), Diffusion Tensor Images (DTI), and Dynamic Susceptibility Contrast (DSC) images are used to predict the response to treatment. Apparent Diffusion Coefficient (ADC) and Fractional Anisotropy (FA) maps are calculated from the DTI data. Relative peak height (rPH) and relative percentage of signal intensity to recovery (rPSR) are extracted using DSC images. Histograms are derived from the CE regions of the maps and statistical features are extracted from the histograms. Predictions are done using a logistic regression (LR) algorithm and the Leave One Out Cross Validation (LOOCV) method is applied to evaluate the quality of the prediction models. Using the DTI data, it is found that the median of FA and ADC are capable of predicting the response to treatment with accuracies of 81.8% and 72.7%, respectively (t-test, p-value=0.029, 0.027). Using the DSC images, it is found that rPH is a predictive feature with 87.5% accuracy (t-test, p-value=0.038). Putting the three features (median of ADC and FA and rPH) together, a fully (100%) accurate prediction is achieved (t-test, p-value=0.014). In conclusion, DTI and DSC images have rich information about cellularity and vascularity of the tumor regions in GBM patients and can be helpful in GBM treatment, especially for antiangiogenic therapy.