Author: Christopher L. Deufel, Suman Gautam, William Y. Song 👨🔬
Affiliation: Virginia Commonwealth University, Mayo Clinic 🌍
Purpose: Direction modulated brachytherapy creates anisotropic dose distribution from an isotropic source. This study aims to develop a truncated conditional value at risk optimization algorithm for Direction Modulated Brachytherapy (DMBT) six groove tandem with ovoid (T&O) or ring (T&R) or needles applicators.
Methods: TCVaR builds upon conditional value at risk (CVaR) by excluding the hottest and coldest voxels within a structure from the mean dose calculation, improving DVH metrics. An iterative convex approximation process was employed to enhance the selection of the excluded voxels. To evaluate TCVaR, Data Envelopment analysis (DEA) was performed to assess the impact of parameter choices and compare 3,125 TCVaR-MCO generated plans against commercially produced plans. The algorithm solves TCVaR-MCO instances as linear optimization problems, utilizing parallelized CPUs for efficiency. Additionally, 25 plans were optimized using TCVaR method with the clinical plan serving as the baseline to construct the Pareto-surface. The clinical plan was created in a BrachyVision treatment planning system (BVTPS), with AcurosBV® model-based dose calculation algorithm (MBDCA).
Results: TCVaR outperformed CVaR in DVH metrics, with improvements when parameters 0.99, β=1.5 and N=5 iterations were applied to exclude the hottest and coldest voxels. Compared to BVTPS, TCVaR improved CTVHR D90% by 0.15% and reduced OARs D2cc doses 0.6% for the bladder, 1.1% for the rectum, and 1.7% for the sigmoid. Computational efficiency was demonstrated by generating 3,125 plans in approximately 5 minutes when executed on VCU’s High-Performance Research Computing Core (HPRC) using 25 nodes and 25 cores in parallel with the Gurobi optimizer.
Conclusion: The TCVaR algorithm for HDR BT demonstrated superior DVH metrics compared to both prior convex optimization algorithms as well as BVTPS. Its computational efficiency and robust performance make it a promising tool for primary optimization in treatment planning or as a quality assurance tool for existing optimization approaches.