Predicting late magnetic resonance image changes in glioma patients after proton therapy

Introduction:

Proton radiation therapy is an effective treatment for glioma patients. To exploit its full potential, a better description of regional differences in radiation response within the brain is required. In this study, we present a model for predicting magnetic resonance (MR) image changes in glioma patients that considers the variability in proton relative biological effectiveness (RBE) as well as the regional susceptibility of brain tissue to radiation damage.

Material and methods:

Six glioma patients treated with adjuvant proton radio(chemo)therapy showed contrast enhancement on follow-up T1-weighted MR imaging corresponding to treatment-related changes. Physical dose and linear energy transfer (LET) were obtained from high-precision Monte-Carlo simulations. The periventricular region (PVR) was contoured as a 4 mm expansion around the ventricles. Correlations between the image changes and dose, LET, and the PVR were evaluated voxel-wise in univariable and multivariable logistic regression analyses. Model performance was assessed by the area under the curve (AUC) using leave-one-out cross validation. The tolerance dose TD50, at which 50% of the tissue voxels show toxicity was interpolated from the model and used for RBE estimation.

Results:

The spatial distribution of MR image changes in the brain was highly non-uniform and correlated poorly with the considered predictors in univariable analysis. Multivariable modelling with either only dose and LET or PVR as third predictor revealed AUC values of 0.89 and 0.92, respectively. TD50 decreased with increasing LET and the modelled RBE was found to vary between 0.9 and 2.3. No relevant difference in model performance was found for track- or dose-averaged LET.

Conclusion:

The strong correlation of MR image changes with dose, LET, and the PVR demonstrates the relevance of non-uniform dose response models. A larger patient cohort including patients without image changes will be used to validate the clinically observed indication of a variable proton RBE.