Assessment of DVH Prediction Model and Auto-Planning Module for Head and Neck VMAT Planning

Abstract

The Auto-Planning module on a treatment planning system (TPS) can efficiently produce clinically acceptable VMAT plans with consistent quality. However, the initial optimization objectives of organs at risk (OARs) in Auto-Planning module need to be manually pre-set, with the result that plan quality and planning time are planner dependent. In this study, a dose volume histogram (DVH) prediction model was implemented for predicting reasonable initial optimization objectives to the Auto-Planning module. The advantage of this combination method was investigated for head and neck VMAT planning. 20 patients with head and neck cancer were selected randomly. 70 Gy in 35 fractions was given to PGTV (planning target volume expanded form gross tumor volume) and 63 Gy in 35 fractions was given to PCTV1 (covering adjacent high-risk structures and bilateral retropharyngeal lymph nodes), with (11 cases) or without (9 cases) 56 Gy in 35 fractions to PCTV2 (for prophylactic lymph nodes if clinically indicated). A DVH prediction model, which uses the geometry of targets and OARs to predict the DVH for OARs in PlanIQ software, was implemented for predicting the initial optimization objectives to the Auto-Planning module. To evaluate the advantage of this combination method, two types VMAT plans were generated as followed: VMATap plans were optimized by the Auto-Planning module with initial objectives from the Auto-Planning template with manually pre-setting objectives, and VMATpre plans were optimized by the Auto-Planning module with initial objectives predicted from the DVH prediction model in PlanIQ. Then, the dosimetric indices between VMATap and VMATpre plans were compared. There were no significant differences of V70Gy, V63 Gy, and V56 Gy between VMATap plans and VMATpre plans for PGTV, PCTV1, and PCTV2, respectively. Additionally, there were no significant differences of Dmax between VMATap plans and VMATpre plans for eye-L, lens-L and optic nerve-L. In some extent, VMATpre plans were more sparing eye-R and len-R than VMATap plans, while VMATap plans produced lower optic nerve-R dose than VMATpre plans. Furthermore, Dmax of brainstem and spinal-cord were both dramatically reduced in VMATpre plans than VMATap plans with significant differences (p < 0.01). Especially for brainstem, Dmax of VMATpre plans were well consistent with the predicted values by the DVH prediction model. For Dmean of parotids, no significant differences were found between these VMAT plans. The reason maybe that the predicted values, initial objectives in VMATpre plans, were around the manually pre-setting initial objectives in VMATap plans. DVH prediction model can generate reasonable initial objectives for the Auto-Planning module in a treatment planning system. The combination method has the potential advantage to further reduce radiation dose received by OARs.