A radiation response model to assess clinical RBE variability in proton therapy

Currently, the clinical implementation of a variable relative biological effectiveness (RBE) in proton therapy is controversially discussed. First clinical evidence indicated a variable RBE for brain irradiation, which needs to be substantiated. For assessing clinical RBE variability, we established a normal tissue response model and applied it to follow-up magnetic resonance (MR) images.

Four glioma patients (grade II-III) showing late morphological T1-weighted contrast-enhanced (T1w-CE) MR image changes and suspicious necrosis were considered. All were treated with passive scattering at the UniversityProtonTherapyDresden (UPTD). Dose, and linear energy transfer (LET) were calculated with a TOPAS-based Monte-Carlo (MC) simulation framework dosimetrically validated for UPTD. To establish radiation response, logistic regression models based on dose and / or LET were trained on T1w-CE MR voxels classified as change (1) or no change (0). Model performance was assessed by the area under the curve (AUC) performing leave-one-out cross validation.

Correlating image changes with dose and LET resulted in high predictive power (AUC=0.87). TD50 values (dose at which 50% of patient voxels show toxicity) decreased linearly with LET (intercept and slope of 88.9 Gy and -10.9 Gy/(keV/μm), respectively). Models considering either dose or LET performed only moderately (AUC of 0.68 and 0.64, respectively). LET averaging method (dose-averaged or track-averaged) had no impact on model performance.

Only the model based on dose and LET led to high predictive power for late MR image changes, suggesting a variable radiation response and, hence, non-constant RBE. This study enables and encourages in-depth assessment of clinical RBE variability in proton therapy.