Click here to


Are you sure ?

Yes, do it No, cancel

Can the Student Outperform Its Master? A Comparison of Two Automated Inverse Planning Engines Using a Quantitative Quality Plan Metric

A Smith1*, A Granatowicz2 , C Stoltenberg1 , S Wang1 , X Liang3 , S Zhou1 , C Enke1 , A Wahl1 , D Zheng1 , (1) University of Nebraska Medical Center, Omaha, NE, (2) Nebraska Medicine, Omaha, NE, (3) University of Florida, Jacksonville, FL


(Saturday, 3/30/2019)  

Room: Exhibit Hall

Purpose: Auto-Planningᵀᴹ and RapidPlanᵀᴹ are two commercial planning engines to automate inverse planning and are based on fundamentally different
methods: Auto-Planning uses a generic template of dose goals to iteratively but automatically modify the planning structures and objectives for plan optimization, and RapidPlan generates static objectives from DVH estimations predicted by models configured using a library of “training� plans. This study objectively compared the performance of these two algorithms on IMRT planning for prostate fossa and lymph nodes adopting the plan quality metric (PQM) used in the 2011 AAMD Plan-Challenge.

Methods: All plans used an identical 9-coplanar-IMRT-beam setup and the 56Gy/68Gy simultaneous-integrated-boost prescribed by the Plan-Challenge. Auto-Planning was used without manual intervention on 20 post-prostatectomy patients and these 20 plans were subsequently employed as the library to build the RapidPlan model. To compare the two engines, an independent test-set of 10 patients plus the Plan-Challenge patient were planned by both Auto-Planning (master) and RapidPlan (student) without manual intervention and evaluated using the PQM that included 14 quantitative plan metrics ranging over target coverage, spillage, and OAR dose. PQM scores were compared between the two engines using the Mann-Whitney U-test.

Results: There was no significant difference between the performances of the two engines on the 11 test plans (p=0.764). On the Plan-Challenge patient, Auto-Planning scored 133 and RapidPlan scored 130.3, as compared with the average human-plan score of 116.9±16.4 (range: 58.2-142.5) among the 125 AAMD Plan-Challenge participants.

Conclusion: Using an innovative study design, an objective comparison has been made between two major commercial automated inverse planning engines. The two engines performed comparably with each other and both yielded plans on par with average human planners. Using a constant-performing planner (Auto-Planning) to train and to compare, RapidPlan was found to yield plans no better than, but as good as, its library plans.

Funding Support, Disclosures, and Conflict of Interest: We would like to acknowledge the research license support of Plan IQ by Sun Nuclear and Auto-Planning by Phillips. We thank Dr. Nelms from ProKnow for some helpful discussion.

Contact Email