Validating knowledge-based volumetric modulated arc therapy plans with a multi-institution model (broad model) using a complete open-loop dataset for prostate cancer.

Abstract

This study examined the characteristics of the broad model (KBP_broad) through a complete open-loop evaluation of volumetric modulated arc therapy (VMAT) plans for prostate cancer in 30 patients at two institutions. KBP_broad, trained using 561 prostate cancer VMAT plans from five institutions with different treatment protocols, was shared with two institutions. The institutions were not involved in the creation of KBP_broad. Plan created with KBP_broad were compared with clinical plans (CPs) and plans created using a single-institution model at each institution (KBP_onsite). KBP_broad maintained the target coverage of CPs while meeting dose limits across varied settings at each institution. At institution X, KBP_broad provided 40, 60, and 70 Gy (V_40Gy, V_60Gy, and V_70Gy, respectively) to 30.8% ± 9.9%, 15.3% ± 8.5%, and 9.0% ± 6.4% of the volume at the rectal wall, respectively, which were significantly smaller than those provided by KBP_onsite and CPs. At institution Y, compared with CPs, KBP_broad provided significantly greater V_50Gy, V_70Gy, dose to 2% of the volume (D_2%) at the rectum, and D_2% at the bladder but significantly lower V_50Gy and V_70Gy at the bladder, in addition to superior dose homogeneity and conformality at the planning target volume. Our complete open-loop evaluation of VMAT plans for prostate cancer at two institutions demonstrated the clinical effectiveness of KBP_broad at institutions producing plans with insufficient reductions in OAR doses. Thus, the quality of KBP_broad plans is likely greater than that of KBP_onsite plans and CPs.