Program Information
Robustness Versus Plan Quality for IMPT
BL Gorissen, N Depauw, J Unkelbach, Massachusetts General Hospital and Harvard Medical School, Boston, MA
Presentations
SU-E-T-684 (Sunday, July 12, 2015) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose:
Range and set-up uncertainties in intensity-modulated proton therapy cannot be addressed with margins. Instead, several robust optimization (RO) models have been developed that explicitly take uncertainty into account. RO optimizes the worst case at the cost of the treatment outcome in more likely scenarios. The goal is to provide insight in the trade-off between robustness and plan quality and the behavior of different trade-off methods.
Methods:
We present four methods to trade-off robustness with plan quality that can be applied to any of the worst case methods found in literature. Each trade-off is a mixture between worst case, expected value, and non-robust planning, which effectively corresponds to assigning different importance weights to error scenarios. Each method is tested on several worst case methods for a sarcoma case, a paraspinal case and a benchmark case from literature.
Results:
Each trade-off method yields a unique dose distribution in the border region between target volume and adjacent normal tissues, corresponding to a specific trade-off between robustness and nominal plan quality. The fully robust solutions perform badly when the realized errors are smaller than the maximum projected errors. Compared to fully robust solutions, significant improvements are possible for non-extreme scenarios while only slightly deteriorating plan quality at an extreme scenario. It is further observed that trade-off methods cannot be mimicked by putting different weights on objectives for the tumor and the normal tissue.
Conclusion:
The method to trade-off robustness with plan quality should be chosen carefully, as each method has a different impact on plan quality. Two of the methods can directly be implemented in any framework for multicriteria optimization, hopefully leading to their quick dissemination.
Contact Email: