Presenter(s)
C. Liu1, and Y. Li2; 1National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital & Shenzhen Hospital,Chinese Academy of Medical Sciences and Peking Union Medical College, Shenzhen, ShenZhen, China, 2Chinese Academy of Medical Sciences Cancer Hospital, Shenzhen Center, Shenzhen, China
Purpose/Objective(s): Proton arc therapy (PAT) achieves high dose conformity for complex tumors via continuous gantry rotation and multi-beam directions, but its optimization is hindered by massive spots, energy layers and beam-angle combinations, leading to long computation time and excessive delivery energy switching. Traditional frameworks fail to balance target dose coverage, OAR sparing and treatment efficiency, prioritizing dosimetric quality over delivery and computational efficiency, hence a unified optimization strategy integrating plan quality, sparsity and delivery speed is needed.
Materials/Methods: A unified PAT optimization framework with a novel iterative algorithm is proposed. Unlike conventional frameworks lacking integrated spot, beam and energy layer sparsification and neglecting energy layer sequencing for delivery efficiency, this framework achieves unified sparsification of the three within a single optimization loop. Centered on a column generation-inspired strategy, it dynamically optimizes the delivery structure by alternating a restricted master problem and a pricing subproblem: the former optimizes spot intensities of current active energy layers via the L-BFGS-B algorithm, and the latter assesses the "reduced cost" of inactive ones. It calculates the gradient of the unified objective function with a regularization term for candidate spots to identify and activate the steepest descent energy layers, and embeds an adaptive pruning mechanism to eliminate redundant layers, enforce delivery efficiency constraints and meet the maximum allowed energy layers per beam. This approach balances dosimetric fidelity and delivery efficiency, generating clinically deliverable plans to maximize target coverage and minimize treatment time. Plans were evaluated by dosimetric metrics (CTV D95%/D98%/D2%, HI, brain stem Dmax/Dmean) and delivery efficiency metrics (ELST, total active energy layers, average layers per beam).
Results: Eight brain tumor cases were included, with the proposed method compared to Gu (2020) baseline method. All plans were normalized to CTV D95% with 56 Gy prescription dose. Dosimetric evaluations showed no significant differences in target coverage and homogeneity between the two methods: the proposed method achieved CTV D98% (55.4±0.3), D2% (60.1±2.6) and HI (1.06±0.04), comparable to the baseline. For OAR sparing, brain stem Dmax had no notable discrepancies. In terms of delivery efficiency, the framework reduced ELST by 38.6% (127±35s vs. 207±17s, p<0.01) and active beams by 41.6% (42 vs. 72), while increasing average active energy layers per beam (2.2 vs. 1). Redistributing energy layers across fewer beams minimized redundant energy switching, cutting ELST without compromising dosimetric quality.
Conclusion: The proposed unified optimization framework shortens PAT energy switching time significantly without sacrificing dosimetric metrics, and has potential clinical value for reducing PAT treatment duration.