Abstract ID: 145 Monte Carlo for CyberKnife radiosurgery with the InCise Multileaf Collimator

Abstract

This work investigates the accuracy and efficiency of Monte Carlo dose calculation with the InciseTM 2 Multileaf Collimator (MLC) for the CyberKnife® M6TM System in Accuray PrecisionTM Treatment Planning System 1.1. Phase space above the MLC is derived from a virtual source model calculated from beam commissioning measurements (profiles, TPRs, and output factors). Transport through the MLC is performed by ray tracing, accounting for leaf tip geometry, leaf-bank tilt, beam hardening, and shifts applied by the MLC controller to compensate for variable offset between the radiographic and geometric beam edge at each leaf tip. Patient model transport uses material and density specific photon mean-free-paths in combination with track-repeating of precomputed electron tracks from mono-energetic photon beams in water, modified using local density. Variance reduction techniques include forced interaction, particle splitting, and range rejection. This version improves on previous implementations by linearly propagating electrons below a cutoff energy in some circumstances instead of depositing their remaining energy in one voxel. The algorithm is implemented in C++, multithreaded over 12 CPU cores, and can be used pre- or post- beam weight optimization. Calculations were compared against measurement for single beams and composed plans in homogeneous and inhomogeneous phantoms. Single beam measurements included central axis percent depth dose (PDD) using diode and microdiamond detectors, and planar dose distributions with EBT3 film. For composed plans, point measurements were performed using air-filled micro-ionization chambers and TLD. Calculation speed was characterized by evaluating lung SBRT plans consisting of between 13 and 126 treatment beams, and PTV’s between 14cc and 72cc. The PDD curves showed good agreement in solid water and water-lung-water phantoms between calculation and measurement. Films had ⩾90% pixels meeting a 2%/2 mm gamma criterion for all 12 aperture shapes in similar phantoms. Composite plans had point dose differences ⩽3% in heterogeneous lung phantoms. Calculation speeds at native 512 × 512 CT resolution (approximately 1 mm3 voxels) and 2% requested uncertainty were 9 min on average, reducing to 3 min when medium resolution (approximately 4 mm3 voxels) was used. Methods to reduce calculation times are under investigation. Preliminary results suggest possible 40%–75% reductions without underlying algorithm or hardware changes.