HDR brachytherapy optimization using simulated annealing coupled with a gradient technique

Several algorithms have been developed for HDR Brachytherapy to optimize dose distributions by determining the source positions and dwell times to minimize user-defined objective functions. The objective functions employed are constrained by physician prescriptions and clinical prerequisites. We present a modified simulated annealing methodology (SA), applied to achieve optimal distributions for some clinical situations. The SA is coupled with a down-hill simplex (gradient) technique to handle dose gradient more efficiently. The gradient technique reduces computing time by as much as 25% over SA alone. This methodology allows great flexibility with regard to a defined hyperdose sleeve, dose to critical structures outside the target volume, and dose homogeneity within the target volume. Results from the aforementioned algorithms are compared to distributions generated using the Nucletron optimization algorithms (regressional analysis), and the Paris System. Breast, anal canal, and nasopharyngeal implant distributions are presented, comparing the three dosimetric approaches. This optimization methodology is at present under development and is not used for routine clinical application