Auto- versus human-driven plan in mediastinal Hodgkin lymphoma radiation treatment

Background: Technological advances in Hodgkin lymphoma (HL) radiation therapy (RT) by high conformal treatments potentially increase control over organs-at-risk (OARs) dose distribution. However, plan optimization remains a time-consuming task with great operator dependent variability. Purpose of the present study was to devise a fully automated pipeline based on the Pinnacle Auto-Planning (AP) algorithm for treating female supradiaphragmatic HL (SHL) patients. Methods: CT-scans of 10 female patients with SHL were considered. A "butterfly" (BF) volumetric modulated arc therapy was optimized using SmartArc module integrated in Pinnacle v. 9.10 using Collapsed Cone Convolution Superposition algorithm (30 Gy in 20 fractions). Human-driven (Manual-BF) and AP-BF optimization plans were generated. For AP, an optimization objective list of Planning Target Volume (PTV)/OAR clinical goals was first implemented, starting from a subset of 5 patients used for algorithm training. This list was then tested on the remaining 5 patients (validation set). In addition to the BF technique, the AP engine was applied to a 2 coplanar disjointed arc (AP-ARC) technique using the same objective list. For plan evaluation, dose-volume-histograms of PTVs and OARs were extracted; homogeneity and conformity indices (HI and CI), OARs dose-volume metrics and odds for different toxicity endpoints were computed. Non-parametric Friedman and Dunn tests were used to identify significant differences between groups. Results: A single AP objective list for SHL was obtained. Compared to the manual plan, both AP-plans offer comparable CIs while AP-ARC also achieved comparable HIs. All plans fulfilled the clinical dose criteria set for OARs: both AP solutions performed at least as good as Manual-BF plan. In particular, AP-ARC outperformed AP-BF in terms of heart sparing involving a lower risk of coronary events and radiation-induced lung fibrosis. Hands-on planning time decreased by a factor of 10 using AP on average. Conclusions: Despite the high interpatient PTV (size and position) variability, it was possible to set a standard SHL AP optimization list with a high level of generalizability. Using the implemented list, the AP module was able to limit OAR doses, producing clinically acceptable plans with stable quality without additional user input. Overall, the AP engine associated to the arc technique represents the best option for SHL.