Advanced kernel methods vs. Monte Carlo-based dose calculation for high energy photon beams

Abstract

Purpose The aim of this study was to compare the dose calculation accuracy of advanced kernel-based methods and Monte Carlo algorithms in commercially available treatment planning systems. Materials and methods Following dose calculation algorithms and treatment planning (TPS) systems were compared: the collapsed cone (CC) convolution algorithm available in Oncentra Masterplan, the XVMC Monte Carlo algorithm implemented in iPlan and Monaco, and the analytical anisotropic algorithm (AAA) implemented in Eclipse. Measurements were performed with a calibrated ionization chamber and radiochromic EBT type films in a homogenous polystyrene phantom and in heterogeneous lung phantoms. Single beam tests, conformal treatment plans and IMRT plans were validated. Dosimetric evaluations included absolute dose measurements, 1D γ-evaluation of depth–dose curves and profiles using 2 mm and 2% dose difference criteria for single beam tests, and γ-evaluation of axial planes for composite treatment plans applying 3 mm and 3% dose difference criteria. Results Absolute dosimetry revealed no large differences between MC and advanced kernel dose calculations. 1D γ-evaluation showed significant discrepancies between depth–dose curves in different phantom geometries. For the CC algorithm γ mean values were 0.90 ± 0.74 vs. 0.43 ± 0.41 in heterogeneous vs. homogeneous conditions and for the AAA γ mean values were 1.13 ± 0.91 vs. 0.41 ± 0.28, respectively. In general, 1D γ results obtained with both MC TPS were similar in both phantoms and on average equal to 0.5 both for profiles and depth–dose curves. The results obtained with the CC algorithm in heterogeneous phantoms were slightly better in comparison to the AAA algorithm. The 2D γ-evaluation results of IMRT plans and four-field plans showed smaller mean γ-values for MC dose calculations compared to the advanced kernel algorithms ( γ mean for four-field plan and IMRT obtained with Monaco MC were 0.28 and 0.5, respectively, vs. 0.40 and 0.54 for the AAA). Conclusion All TPS investigated in this study demonstrated accurate dose calculation in homogenous and heterogeneous phantoms. Commercially available TPS with Monte Carlo option performed best in heterogeneous phantoms. However, the difference between the CC and the MC algorithms was found to be small.