Click here to

Session: Therapy: External Beam: Automatic Treatment Planning [Return to Session]

A Feasibility Study of Automated Inverse Planning Using Hyperparameter Tuning

M Kim*, K Maass, A Aravkin, University of Washington, Seattle, WA


MO-IePD-TRACK 5-4 (Monday, 7/26/2021) 12:30 PM - 1:00 PM [Eastern Time (GMT-4)]

Purpose: Inverse planning in current practice requires treatment planners to modify multiple parameters in the objective function to produce clinically acceptable plans. Due to the manual steps in this process, the quality of plans could widely vary depending on planning time available and treatment planner’s skills. The purpose of this study is to automate the inverse planning process to maintain plan quality with reduced planning time.

Methods: The parameters for tuning were the limit dose parameters of each organ-at-risk (OAR) objective. For given parameters, the treatment plan optimization was done in Raystation using the scripting interface to obtain the dose distributions deliverable. We investigated random and Bayesian searches with various forms of utility functions to achieve clinical goals. We also varied the range of dose parameters in the search space. We applied the framework developed to 5 patients who received SBRT to their peripherally located lung tumors with distinct clinical concerns. We constrained the dose to OARs according to RTOG0915. We normalized all plans to have PTV D95 equal the prescription dose and compared the OAR dose metrics in the clinically used plans with those in the automatically generated plans.

Results: The number of iterations required to produce the satisfactory plan quality was 100 and the average planning time was 2.2 hours. The OAR doses achieved by the automatically generated plans were lower than the clinical plans by up to 76.8%. When the OAR doses were larger than the clinical plans, they were still less than the clinical goals by 3.6-98.9% indicating they were clinically acceptable.

Conclusion: We successfully devised an automated inverse planning framework using hyperparameter tuning. We demonstrated that the plan quality could be improved, and the average treatment planning time (no planner intervention required) was 2.2 hours for SBRT lung cases.



    Not Applicable / None Entered.


    Not Applicable / None Entered.

    Contact Email