Encrypted login | home

Program Information

Multicriteria Optimization for Brachytherapy Treatment Planning

V Wu

V Wu1*, M Epelman1 , M Sir2 , K Pasupathy2 , M Herman2 , C Deufel2 , (1) University of Michigan, Ann Arbor, MI, (2) Mayo Clinic, Rochester, MN


WE-AB-605-9 (Wednesday, August 2, 2017) 7:30 AM - 9:30 AM Room: 605

Purpose: Delays in the brachytherapy treatment planning (TP) process are associated with increased anesthesia use, patient discomfort, and risk of treatment delivery inaccuracy, which may worsen toxicity or disease control. Current commercial brachytherapy TP systems use an iterative ‘guess-and-check’ approach that is inefficient for exploring trade-offs in dose between targets and healthy structures. This work introduces an efficient Pareto-style planning approach and intuitive GUI that enables a planner or physician to directly explore DVH trade-offs without iteratively re-optimizing.

Methods: Plans were optimized using a “truncated” conditional value-at-risk (CVaR, a mean tail dose) to approximate intrinsic DVH metrics in a convex multi-criteria optimization (MCO) framework. The MCO problem was solved repeatedly by varying the bounds on criteria represented as constraints and optimizing one criterion to generate a library of high-quality candidate plans. By interpolating the dwell times from the library plans, we efficiently increased library density for a trade-off surface (represented via GUI) without additional re-optimization and maintained deliverability. Process feasibility was verified by planning retrospectively cervix, prostate, and breast treatment sites.

Results: Plans obtained using a truncated CVaR approximation had generally superior DVH results than plans from a DVH-based commercial TP system. Choice of CVaR tail and truncation sizes influenced the quality of the approximation. Truncation improved CVaR approximations of upper tail DVH metrics (e.g., PTV D1cc). Average candidate plan generation time was <30 seconds. Linearly-interpolated dwell times of candidate plans result in typically <0.5% loss in DVH metrics quality.

Conclusion: The brachytherapy treatment planning approach efficiently generates a trade-off surface consisting of high-quality plans that span a wide range of DVH values for each structure of interest. The method is generalizable to any number of criteria, and library generation is trivially parallelizable. Represented as an intuitive GUI, this tool could improve both TP time and quality for brachytherapy.

Contact Email: