(Sunday, 7/14/2019) 4:00 PM - 5:00 PM
Room: Stars at Night Ballroom 4
Purpose: To develop a computationally efficient algorithm to handle dose-volume constraints (DVCs) for IMRT fluence map optimization without any parameter tweaking and incorporate that in our in-house automated treatment planning system.
Methods: DVCs result in a computationally intractable non-convex optimization problem. We proposed a 2-phase approach, wherein phase-1, we solved an optimization problem including convex relaxation of DVCs. Although this convex relaxation did not guarantee DVC satisfaction, it provided crucial initial information about voxels that received low-dose radiation. Subsequently, phase-2 solved an optimization problem with maximum dose constraints imposed on low-dose regions. We categorized DVCs into hard and soft DVCs, where hard DVCs were strictly enforced by the optimization and soft DVCs were allowed to be violated. However, the violations were penalized in the objective function. The approach was tested on a series of paraspinal, lung and oligomet cases. The algorithm was also compared against the ground-truth solution obtained by solving a non-convex optimization problem using mixed integer programming.
Results: The proposed algorithm successfully met all the hard DVCs without significantly compromising the PTV coverage. For soft DVC, the DVH curve moved toward the desired direction with an insignificant degradation of PTV coverage. For a low-resolution paraspinal case, we obtained the ground-truth solution in 15 hrs while the proposed algorithm converged in only 1 min. The overall computational time was increased by 20%, 6% and 7% for paraspinal, lung and oligomet patient respectively with the DVC as compared to the algorithm without the DVC.
Conclusion: A novel computationally tractable algorithm to handle hard and soft DVCs was developed which was capable of satisfying DVCs without any parameter tweaking. Although the algorithm was adopted by our in-house developed automated treatment planning system, it can in principle be used in any constrained optimization framework.
Funding Support, Disclosures, and Conflict of Interest: This work was partially supported by the MSK Cancer Support Grant/Core Grant (P30 CA008748), and the Enid A. Haupt Endowed Chair Fund.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
