Background and purpose: RapidPlan can be used to extract patient’s anatomy and dose information from intensity-modulated radiotherapy (IMRT) plans to predict the dose-volume histogram (DVH) of a new one

in the method of DVH estimation models. Establishing a knowledge-based model for treatment planning system

(TPS) and accelerator separately requires a lot of efforts

and the selection will be cumbersome. This study aimed to investigate whether a knowledge-based treatment planning model can smoothly migrate to different machine and TPS. Methods: The clinical treatment plans of 50 cervical cancer patients treated in Fudan University Shanghai Cancer Center from 2015 to 2016 were added in RapidPlan to develop a knowledge-based planning model. All training data were created with Pinnacle and optimized for 6 MV photon beams from a Synergy accelerator. Model was used to estimate the DVH in 15 IMRT plans. Plans were reoptimized to evaluate the impact of the accelerator model and TPS on knowledge-based planning model after extracting the objective function value. The evaluation included 3 groups. In Group 1

the knowledge based plan (KBP) and manual plan used the same accelerator and TPS as model training data (Pinnacle and Synergy). In Group 2

the KBP and manual plan used the same TPS

while the accelerator model was different (Pinnacle and Truebeam). In Group 3

the KBP and manual plan used different TPS and accelerator models (Eclipse and Truebeam). DVH quantitative analysis was performed to make comparison between the KBP and the manual plans in 3 groups respectively. Results: In Group 2 and Group 3

KBP plans showed similar quality of planning target volume (PTV) as manual plans. However

KBP plans improved D

2

% (0.95 Gy

P0.01) and HI (0.02

P0.01) in Group 1. RapidPlan decreased the average values of V

30

V

45

and mean dose of bladder in all 3 groups. RapidPlan also generated better mean dose of bowel in 3 groups. Conclusion: KBP does not significantly depend on machine and TPS .