Dosimetric comparison of Pinnacle3 9.2, Eclipse(TM) 11.0, and iPlan 4.1 algorithms with heterogeneous phantoms

Our goal is to compare the dosimetric accuracy of the Eclipse(TM) 11.0 Anisotropic Analytical Algorithm (AAA) and AcurosRTM XB (AXB), Pinnacle3 9.2 Collapsed Cone Convolution, and the iPlan 4.1 Monte Carlo (MC) and Pencil Beam (PB) algorithms using measurement as the gold standard.;Measurements were taken for a range of beam energies beams and field sizes in heterogeneous block phantoms and anthropomorphic phantom, which mimics lung, spine, ribs, and tissue. Heterogeneous interfaces and buildup regions were investigated in the block phantom setups. The measurements were taken using an ion chamber and diode detectors. Dose in a humanoid phantom at our patient target volume and out of field regions wre investigated in the anthropomorphic phantom. Data from the planning systems were computed for each scenario, and compared to our benchmark measurements using percent differences.;The best results between data from the algorithms and our measurements occur after the buildup region in solid water for block phantoms, and at the treatment isocenter in anthropomorphic phantom. The cumulative results from the block phantoms (using a 3% difference from benchmark as our passing criteria) indicate that in general, AXB performs the best, but in some regions other algorithms excel over AXB. iPlan MC performs better at points outside the beam. AAA performs as well as AXB at the lung interfaces. MC and AAA perform best in the buildup regions. All algorithms except PB are within 5% difference from our benchmark measurements at the isocenter. Differences between our measurements and algorithm data are much greater for the off-axis points.;In summary, there is an obvious lack of accuracy for doses to critical structures outside any primary beam. Further studies will need to be done to understand the risk of dose to OARs outside the primary fields. The data for the anthropomorphic phantom indicates that all the algorithms accurately compute dose to targets within lung, excluding PB. Finally, our study displays the possible power behind AXB, overall outperforming the other algorithms.