MO‐D‐BRB‐08: Non‐Voxel‐Based‐Broad‐Beam (NVBB) Framework for IMRT Treatment Planning

Purpose IMRT treatment planning, as implemented in the TomoTherapy TPS, is a very‐large‐scale‐problem (VLSP) and requires a computer cluster. We present a novel framework that enables IMRT planning solvable in a single workstation with improvements in cost, plan quality and planning throughput. Material and Methods The current TomoTherapy TPS uses a beamlet‐based‐optimization that requires pre‐calculation and storage of large amount of beamlets. A computer cluster with 7–14 nodes is used for both computation and data parallelization. We invented and developed a Non‐Voxel‐based‐Broad‐Beam (NVBB) framework for direct‐machine‐parameter‐optimization (DMPO) in IMRT. Both functions and derivatives are evaluated in the continuous viewpoint without “voxel” and “beamlet” representation. Thus pre‐calculation and storage of beamlets are no longer needed. Low‐memory, full computation and data parallelization nature of the framework render its efficient implementation on the graphic‐procession‐unit (GPU). We implemented the NVBB framework in TomoTherapy TPS (NVBB‐TPS). NVBB‐TPS runs on a single workstation with a GPU. Significant verification/validation tests were performed in house and via third parties. Benchmarks on dose accuracy, plan quality and throughput were compared with the commercial TomoTherapy TPS (cluster‐TPS) using 14 nodes. Results Compared with the cluster‐TPS, the preprocessing time of NVBB‐TPS reduced from 10–200 minutes to about 10 seconds and the time of each iteration is reduced to 25–80%. Plan qualities were indistinguishable for most cases and less dose artifacts were present for a few cases via NVBB‐TPS. For the same delivery plan, dose differences between cluster‐TPS and NVBB‐TPS were within 1%, 1mm for all test cases. Conclusions We developed a novel NVBB framework for VLSP IMRT optimization. The DMPO nature of the algorithm eliminates the needs for beamlets and leads to better plan quality. Non‐cluster solution significantly saves hardware and service costs. Compared with the cluster‐TPS, treatment time can be reduced in many folds with NVBB‐TPS.