Evaluation of Knowledge-Based Planning Model In Head and Neck Cancer Using Volumetric Modulated Arc Therapy on a Halcyon System
Abstract
Purpose
RapidPlan (RP) was utilized to potentially reduce variability of plan quality and decrease optimization times. When implementing a new treatment delivery system such as the Halcyon, planners may require time to establish a dataset suitable for training a new KBP model. This study aimed to evaluate the performance of a KBP model trained on head and neck cancer treated with a conventional linac using 6 MV flattening filter beams, which can effectively generate volumetric modulated arc therapy (VMAT) plans for the Halcyon with 6 MV flattening filter-free beams.
Methods
The RP model was trained with 60 patients. All plans had been used in previous treatments with a conventional linac with three targets at dose levels of 54/60/70 Gy in 33 fractions. To assess the applicability of the model for Halcyon, an independent cohort of 18 patients not included in the training dataset was used to evaluate. The dosimetric outcomes and planning efficiency of VMAT plans produced on TrueBeam (RP_TB) were compared with those produced on Halcyon (RP_Hal).
Results
The result showed that the RP_TB plans demonstrated significantly higher dose homogeneity in the target at dose 70 Gy compared with the RP_Hal plans. However, RP_Hal plans significantly reduced the mean dose to parallel organs, including left and right parotid glands and oral cavity, compared to RP_TB plans, while the maximum dose to serial organs did not increase significantly. The average planning time with single optimization was less than 30 minutes. The pre-treatment verifications for RP_Hal plans with a gamma index (3%, 2mm) higher than 99% in all cases.
Conclusion
The KBP model achieved greater consistency and enhanced plan quality for head and neck cancer cases. Moreover, the KBP model trained with a conventional linac can provide high-quality VMAT plans for Halcyon, despite differences in beam energy and collimator design.