Impact of Fluence-Based Constraints on VMAT Plan Optimization
Abstract
Purpose
Volumetric-modulated arc therapy (VMAT) plan optimization remains a challenging problem due to its non-convex character and the large dimensionality of the solution space. Commonly, the problem is split in three steps: 1) ideal fluence optimization 2) arc sequencing and 3) direct aperture optimization (DAO). Traditionally, the optimized fluence is only used to generate a good initial solution. Additional utilization through fluence-based regularization in the DAO cost function may yield further benefits. Here we investigate the benefits of such regularizations in terms of organ at risk dose and plan complexity, which is especially relevant for MR-linac implementations.
Methods
We introduce the Trade-off Index (TOI), which encodes a direction in the solution space towards which DAO solutions flow after the introduction of a penalty function. The TOI metric takes values between 0 and 1, with 0 indicating pure benefit and 1 pure loss from the introduction of a penalty. We evaluated four regularization strategies: (1) a quadratic deviation from the ideal fluence; (2) a clustered quadratic fluence-based regularization in which the ideal fluence is first clustered; (3) a geometric constraint that restricts multileaf collimator (MLC) edges to lie within the region defined by the top p% of beamlets in grid space; and (4) a dose-based regularization that penalizes deviations of the effective dose from the ideal fluence dose. Each regularization technique was assessed using the proposed TOI metric in a prostate treatment plan.
Results
The average (standard deviation) TOI for each tested metric was: 0.14(0.04) for the quadratic fluence, 0.19 (0.07) for the clustered fluence, 0.90 (0.23) for the geometric penalty and 0.53 (0.32) for the dose-based constraint.
Conclusion
We provide a standardized workflow to evaluate the benefit of different regularizations in the DAO optimization cost function. This can lead to straightforward decisions on the efficacy of plan optimization constraints.