An iterative sequential mixed-integer approach to automated prostatebrachytherapy treatment plan optimization

Abstract

Conventional treatment planning for interstitial prostate brachytherapy isgenerally a `trial and error' process in which improved treatment plans aregenerated by iteratively changing, via expert judgement, the configuration ofsources within the target volume in order to achieve a satisfactory dosedistribution. We have utilized linear mixed-integer programming (MIP) and thebranch-and-bound method, a deterministic search algorithm, to generatetreatment plans. The rapidity of dose falloff from an interstitial radioactivesource requires fine sampling of the space in which dose is calculated. Thisleads to a large and complex model that is difficult to solve as a single 3Dproblem. We have therefore implemented an iterative sequential approach thatoptimizes pseudo-independent 2D slices to achieve a fine-grid 3D solution.Using our approach, treatment plans can be generated in 20-45 min on a200 MHz processor. A comparison of our approach with the manual `trial anderror' approach shows that the optimized plans are generally superior. Thedose to the urethra and rectum is usually maintained below harmful levelswithout sacrificing target coverage. In the event that the dose to the urethrais undesirably high, we present a refined optimization approach that lowersurethra dose without significant loss in target coverage. An analysis of thesensitivity of the optimized plans to seed misplacement during theimplantation process is also presented that indicates remarkable stability ofthe dose distribution in comparison with manual treatment plans.