Inverse Optimization Using Smoothness Control for Rotating Shield Brachytherapy
Y Liu1, R Flynn2*, Y Kim2, X Wu1,3, (1) Department of Electrical and Computer Engineering, The University of Iowa, Iowa City, IA, (2) University of Iowa Hospitals and Clinics, Iowa City, IA (3) Department of Radiation Oncology, The University of Iowa, Iowa City, IASU-E-T-585 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: To present an optimization method for Rotating-Shield Brachytherapy (RSBT) treatment plans that produces smoothly-varying dwell time distributions, enabling the efficient delivery of smooth dose distributions.
Methods: For each of 5 cervical cancer cases, a beamlet set with fine azimuthal emission angle (11.25 degree) were generated by TG-43 dose calculation model. These beamlets are called baseline beamlets. Delivery plans that use baseline beamlets are called baseline plans. To optimize a baseline plan, a dose optimizer was used to assign dwell times to each beamlet with two objectives balanced: (i) the total dif-ference between the delivered dose and the prescribed dose of the surface voxels is minimized; and (ii) the total variation (TV) of the dwell times is minimized. The optimization was done by solving a quadratic programming (QP) problem. For comparison, a simulated annealing based inverse optimization was performed and the computational time was limited by 5 minutes per case. The impact of the smoothness was investigated by performing dual rotating-shield brachytherapy (D-RSBT) planning with the baseline plans.
Results: The total variation of the dwell times from the QP-generated plans decreased by a factor ranging from 5 to 93 relative to the SA-generated plans. The delivery efficiency curves were improved given delivery time budget limited below ~20 min/fx. With the delivery time specified at 15 min/fx, the high-risk clinical target volume D₉₀ from D-RSBT was improved by 15% on average compared to SA method, ranging from -20% to 30%.
Conclusions: Improving the smoothness of baseline plans potentially improves the delivery quality of RSBT as the dose distribution quality of a smoothed baseline plan is easier to maintain with limited delivery time than an unsmoothed one. Minimizing the total variation in the optimization process increases the smoothness of D-RSBT dose distributions relative to techniques that neglect dwell time smoothness.