Clinical implementation in proton therapy of multi-field optimization by a hybrid method combining conventional PTV with robust optimization

To implement a robust multi-field optimization (MFO) technique compatible with the application of a Monte Carlo (MC) algorithm and to evaluate its robustness. Nine patients (three brain, five head-and-neck, one spine) underwent proton treatment generated by a novel robust MFO technique. A hybrid (hMFO) approach was implemented, planning dose coverage on isotropic PTV compensating for setup errors, whereas range calibration uncertainties are incorporated into PTV robust optimization process. hMFO was compared with single-field optimization (SFO) and full robust multi-field optimization (fMFO), both on the nominal plan and the worst-case scenarios assessed by robustness analysis. The SFO and the fMFO plans were normalized to hMFO on CTV to obtain iso-D95 coverage, and then the organs at risk (OARs) doses were compared. On the same OARs, in the normalized nominal plans the potential impact of variable relative biological effectiveness (RBE) was investigated. hMFO reduces the number of scenarios computed for robust optimization (from twenty-one in fMFO to three), making it practicable with the application of a MC algorithm. After normalizing on D95 CTV coverage, nominal hMFO plans were superior compared to SFO in terms of OARs sparing (p   <  0.01), without significant differences compared to fMFO. The improvement in OAR sparing with hMFO with respect to SFO was preserved in worst-case scenarios (p   <  0.01), confirming that hMFO is as robust as SFO to physical uncertainties, with no significant differences when compared to the worst case scenarios obtained by fMFO. The dose increase on OARs due to variable RBE was comparable to the increase due to physical uncertainties (i.e. 4-5 Gy(RBE)), but without significant differences between these techniques. hMFO allows improving plan quality with respect to SFO, with no significant differences with fMFO and without affecting robustness to setup, range and RBE uncertainties, making clinically feasible the application of MC-based robust optimization.