Click here to

Session: Therapy: External Beam: treatment planning Optimization [Return to Session]

Solving the Non-Convex Volumetric Modulated Arc Therapy (VMAT) Problem Via a Sequence of Convex Optimization Problems

P Dursun*, G Jhanwar, J Deasy, M Zarepisheh, Memorial Sloan Kettering Cancer Center, New York, NY


MO-IePD-TRACK 5-2 (Monday, 7/26/2021) 5:30 PM - 6:00 PM [Eastern Time (GMT-4)]

Purpose: Volumetric modulated arc therapy (VMAT) optimization problem is a complex and highly non-convex problem and existing algorithms suffer from local optimality. We propose a novel sequential convex programming (SCP) algorithm that directly optimizes aperture shapes and promotes delivery efficiency and global optimality.

Methods: We propose an algorithm for VMAT that conducts an effective global search by solving a sequence of convex local approximations. The algorithm starts with optimizing the aperture weights of 72 evenly distributed beams using the beam’s eye view (BEV) of the target from each direction. Thereafter, a convex approximation problem is solved to optimize the leaf positions and the aperture weights within a search space where the leaves can move in and out within a pre-defined step-size. There are both local and global search strategies in the algorithm to ensure a high-quality plan. The complexity of the apertures is constrained, to exclude highly irregular aperture shapes, and is also penalized in the objective function to increase shape regularity. The algorithm is tested on paraspinal, prostate and oligometastasis sites. The VMAT plans are compared against ideal 72-beam IMRT plans. Also, the performance of aperture regularization strategies is analyzed.

Results: The DVHs of the optimized VMAT plans are similar to 72-beam IMRT plans. Due to the convexity of the resultant sub-problems, the algorithm is computationally tractable, converging in less than an hour on average on a regular Windows workstation. Significant reductions in aperture complexity at a reasonable cost of plan quality are observed after adding constraints and the aperture complexity regularization term in the objective function.

Conclusion: We developed a computationally efficient algorithm for VMAT which finds a near-global optimal solution by alternating large (global) and small (local) search steps. Aperture complexity is effectively constrained (hard constraint) and penalized (objective function) to produce efficient delivery plans.



    Optimization, Treatment Planning, Radiation Therapy


    Not Applicable / None Entered.

    Contact Email