Purpose: Respecting luminal OAR constraints while maximizing target coverage is challenging for pancreas SBRT due to adjacency of the structures. Resultant plan quality is dependent on plannerâ€™s experience. To improve plan quality consistency, we hypothesize that optimal fluence can be deduced from previous plans with similar anatomy. This study investigated the feasibility of using atlas matching to initialize and accelerate fluence optimization for pancreas SBRT.
Methods: Thirty patients treated with pancreas SBRT were included. To build the atlas, the contours of PTV and duodenum (OAR) were first extracted. All axial contour slices were categorized into three groups based on the minimum PTV-OAR distance d_min: PTV-only, adjoining/adjacent OAR (d_minâ‰¤10mm), and distant OAR (d_min>10mm). Then, baseline 9-field IMRT plans using uniformly-initialized fluence were generated using an in-house optimizer. To find the queryâ€™s best matched atlas, a scoring system defining anatomy similarity metrics was developed. The atlas caseâ€™s fluence was transferred to the query PTV with scaling and set as the initial fluence. The atlas-guided fluence optimization proceeded with the same constraints for the query case. A leave-one-out cross validation was performed to assess the feasibility of the proposed strategy. The atlas-guided plans were compared with uniformly-initialized plans in terms of cost function values and dosimetric endpoints. Paired t-test was performed.
Results: Atlas-guided fluence optimization reduced the mean initial cost function value by 69.7% (p<0.01). For dosimetric endpoints, the average OAR mean dose and PTV V100% were 14.07Gy and 37.1% for atlas-guided plans, and 14.21Gy (p<0.01) and 35.7% (p<0.01) for uniformly-initialized plans.
Conclusion: Results showed feasibility to use prior fluence from plans with similar anatomical patterns to guide pancreas SBRT fluence optimization. It offered customized initialization which achieved improved dosimetric endpoints. This fluence optimization scheme could potentially reduce inter-planner quality variation.
Funding Support, Disclosures, and Conflict of Interest: This work was partially supported by NIH/NCI 1R01CA201212.