Abstract
PURPOSE: Volumetric-modulated arc therapy (VMAT) treatment planning allows a compromise between a sufficient coverage of the planning target volume (PTV) and a simultaneous sparing of organs-at-risk (OARs). Particularly in the case of lung tumors, deciding whether it is possible or worth spending more time on further improvements of a treatment plan is difficult. Therefore, this work aims to develop a knowledge-based, structure-dependent, automated dose volume histogram (DVH) prediction module for lung tumors. METHODS: The module is based on comparing geometric relationships between the PTV and the surrounding OARs. Therefore, treatment plan and structure data of 106 lung cancer cases, each treated in 28 fractions and 180 cGy/fx, were collected. To access the spatial information, a two-dimensional metric named overlap volume histogram (OVH) was used. Due to the rotational symmetry of the OVH and the typically coplanar setup of the VMAT technique, OVH is complemented by the so-called overlap-z-histogram (OZH). A set of achievable DVHs is predicted by identifying plans in the database with similar OVH and OZH. By splitting the dataset into a test set of 22 patients and a training set of 84 patients, the prediction capability of the OVH-OZH combination was evaluated. For comparison between the predicted and achieved DVH curves the coefficient of determination R(2) was calculated. RESULTS: The total lung showed strong linearity between predicted and achieved DVH curves for the OVH-OZH combination, resulting in a R2 value close to 1 (0.975 ± 0.022). The heart benefits the most of the OZH resulting in a high prediction capability, with a higher R2 of 0.962 ± 0.036 compared to the prediction with OVH only (0.897 ± 0.087). CONCLUSION: The combination of OZH and OVH was suitable for building a knowledge-based automated DVH prediction module. Implementing this method into the clinical workflow of treatment planning will contribute to advancing the quality of VMAT plans.