MultiRBE: Treatment planning for protons with selective radiobiological effectiveness

Clinical treatment planning protocols for protons recommend a uniform value radiobiological effectiveness (RBE) of protons of 1.1 throughout the treatment field, despite evidence from in-vitro and animal studies that proton RBE increases with linear energy transfer (LET), causing tissues placed distally to the target location to receive a presumably higher biological dose than estimated. While several voices in the medical physics community have advocated for variable RBE-based optimization, the uncertainties in RBE models have prevented its implementation in clinical practice, since an overestimation of RBE could cause significant target underdosage.

We propose a mixed RBE model (MultiRBE), where a uniform RBE is used in the target contours to ensure an adequate tumor coverage in terms of physical dose, but a variable RBE is used elsewhere. Our model was implemented in the open-source treatment planning system matRad and three example cases were planned: a homogeneous phantom, a prostate tumor and a head-and-neck case. MultiRBE was used for plan optimization, and the produced plans were subsequently evaluated in terms of physical dose coverage (V_95% ) and variable RBE-weighted dose in organs at risk and normal tissue complication probabilities (NTCP), where prediction models were available.

The planning algorithm showed potential for reducing the biological dose in organs surrounding the planning target and thus decreasing the probability for complications in normal tissue (by up to 62% in the prostate case and 37% in the head-and-neck patient). This was achieved without compromising the target coverage or homogeneity in terms of physical dose, as a result of a smarter redistribution of dose among the surrounding tissues with regard to the optimization constraints.

The results prove the ability of the MultiRBE model to reduce biological dose at healthy tissues without compromising the dose coverage of the tumor, with independence of the variable RBE models used.