SU-E-T-613: Improving Dose-Shaping and OAR-Sparing Using Robust Statistical Methods.

Abstract

To achieve a better balance between tumor coverage and organ-at-risk (OAR) sparing by applying robust statistical methods to IMRT inverse optimization. We propose a novel approach to shape dose dropoff from the PTV by minimizing the L1 norm of the difference between the obtained and aimed dose value and the total variation of the dose distribution on the PTV and OAR regions. Minimizing the L1 norm results in operand values that are mostly small but potentially large on a sparse set. Applying this structure to both dose discrepancy and dose gradient, our method achieves: 1) sparsely-distributed, high-dose gradients; 2) structure dose-homogeneity; and 3) connectivity in the distribution of high-gradient voxels on the PTV's surface. Dose-volume histograms (DVHs) and visual observations of dose maps were compared among two plans optimized for the same dynamic IMRT prostate case: 1) the clinical plan (plan_c); and 2) a plan optimized using the TV energy, L-1 norm, and an over-dose quadratic (plan_r). Importance factors for the objective function of planr were chosen empirically. The beam geometry was obtained from the clinical plan, and the solution was optimized with linear programming. Plan_r improved dose conformity, increased overall bladder-sparing, and globally increased rectum-sparing relative to plan_c. These results suggest that robust formulations may be used to improve dose-shaping and OAR-sparing. The numerical stability and implementation simplicity of our method permits fast translation to the clinic.