Robust Optimization of High Dose-Rate Brachytherapy
Timmy Siauw, Ken Goldberg, Ron Alterovitz and Adam Cunha
Brachytherapy cancer treatment uses preoperative images to create optimized treatment plans. However, there is inherent uncertainty in the contouring of these organs, among other types of uncertainty associated with the position and size of organs. The discrepancies between the contouring decisions of the physician and the actual contours of the organ can make a big difference in the quality of the treatment.
This project aims to capture contour uncertainty in a meaningful way and incorporate it into the HDR optimization. The uncertainty is simulated by assuming the true organ contours lie between an inside and outside margin. We use the dose-rate parameters calculated at these margins to bound the true dose-rates. We can then use this information to robustify the original LP according to the interval methods developed by Ben Tal and Nemirovski.
We would like the solution to the robust linear optimization model to remain stable with respect to medical metrics such as dosimetric indices even under large uncertainties. We evaluated our treatment plan against the nominal one for various simulated deviations from the original contours and checked that our treatment plan does not create excessive radiation “hot spots.”
In future work, we hope to evaluate the accuracy of our uncertainty simulation using real contour data, and quantify the performance of the robust solution in a more realistic setting.
Figure 1: Optimization objective function value versus expansion (dose point uncertainty) for the nominal and robust solutions. The robust objective is more stable.