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

Conventional treatment planning for interstitial prostate brachytherapy is generally a 'trial and error' process in which improved treatment plans are generated by iteratively changing, via expert judgement, the configuration of sources within the target volume in order to achieve a satisfactory dos e distribution. We have utilized linear mixed-integer programming (MIP) and the branch-and-bound method, a deterministic search algorithm, to generate treatment plans. The rapidity of dose falloff from an interstitial radioac tive source requires fine sampling of the space in which dose is calculated . This leads to a large and complex model that is difficult to solve as a s ingle 3D problem. We have therefore implemented an iterative sequential app roach that optimizes pseudo-independent 2D slices to achieve a fine-grid 3D solution. Using our approach, treatment plans can be generated in 20-45 mi n on a 200 MHz processor. A comparison of our approach with the manual 'tri al and error' approach shows that the optimized plans are generally superio r. The dose to the urethra and rectum is usually maintained below harmful l evels without sacrificing target coverage. In the event that the dose to th e urethra is undesirably high, we present a refined optimization approach t hat lowers urethra dose without significant loss in target coverage. An ana lysis of the sensitivity of the optimized plans to seed misplacement during the implantation process is also presented that indicates remarkable stabi lity of the dose distribution in comparison with manual treatment plans.