A comparison of voxel sampling approaches for intensity-modulated radiation therapy.

backgroundIntensity-modulated radiation therapy (IMRT) treatment planning technology has made personalized cancer care a reality for millions of patients in recent decades. The complexity of these tools, however, often leads to intractability and suboptimal mitigation strategies in otherwise well-optimized plans. An example of such a strategy is reducing data granularity through voxel sampling. PURPOSE: While the practice of sampling is nearly universal in treatment planning, methods vary widely across studies and clinical practices, and few studies directly compare these approaches. The goal of this study is to examine the relative quality loss of some of the more common voxel sampling methods proposed in the literature.

methodsFive core sampling approaches identified from the literature, and their variations, were distilled into eight distinct sampling methods. A comparative framework, with multiple fluence map optimization models was developed into a testing pipeline built on MATLAB, C++, and CPLEX. This pipeline was then used to evaluate the trade-off between computational complexity and information loss across all sampling approaches. The pipeline was run on the open-source CORT dataset as well as a retrospective clinical lung, prostate, chest wall, esophagus and neck studies, at sampling rates that disregarded roughly 87.5%, 93.7%, and 96.9% of patient voxels, respectively.

resultsOf the eight methods tested, coordinate-based

conclusionsThe proposed framework was successfully able to inform the selection of sampling approach based on patient data characteristics in the IMRT setting. Future extensions could lead to enhanced sampling methods as well as methodological expansions to more radiation treatment modalities.