Commissioning of GPU–Accelerated Monte Carlo Code FRED for Clinical Applications in Proton Therapy

We present commissioning and validation of Fred, a graphical processing unit (GPU)–accelerated Monte Carlo code, for two proton beam therapy facilities of different beam line design: CCB (Krakow, IBA) and EMORY (Atlanta, Varian). We followed clinical acceptance tests required to approve the certified treatment planning system for clinical use. We implemented an automated and efficient procedure to build a parameter library characterizing the clinical proton pencil beam. Beam energy, energy spread, lateral propagation model, and a dosimetric calibration factor were parametrized based on measurements performed during the facility start-up. The Fred beam model was validated against commissioning and supplementary measurements performed with and without range shifter. We obtained 1) submillimeter agreement of Bragg peak shapes in water and lateral beam profiles in air and slab phantoms, 2) (Formula presented.) dose agreement for spread out Bragg peaks of different ranges, 3) average gamma index (2%/2 mm) passing rate of (Formula presented.) for (Formula presented.) patient verification measurements using a two-dimensional array of ionization chambers, and 4) gamma index passing rate of (Formula presented.) for three-dimensional dose distributions computed with Fred and measured with an array of ionization chambers behind an anthropomorphic phantom. The results of example treatment planning study on (Formula presented.) patients demonstrated that Fred simulations in computed tomography enable an accurate prediction of dose distribution in patient and application of Fred as second patient quality assurance tool. Computation of a patient treatment in a CT using (Formula presented.) protons per pencil beam took on average 2′30 min with a tracking rate of 2.9 (Formula presented.) (Formula presented.) (Formula presented.). Fred was successfully commissioned and validated against the clinical beam model, showing that it could potentially be used in clinical routine. Thanks to high computational performance due to GPU acceleration and an automated beam model implementation method, the application of Fred is now possible for research or quality assurance purposes in most of the proton facilities.