Purpose: To compare clinical dosimetry calculated by two advanced algorithms, each used in a mainstream treatment planning system, for LINAC-based brain SRS.
Methods: We evaluated 138 plans of 108 consecutive patients, with 443 targets (both primary and metastatic), receiving LINAC-based (with high-definition multileaf-collimators) single- or multiple-fraction brain SRS planned using a fast-Monte-Carlo algorithm (FMC) in BrainLab Elements. These included both single-target volume-modulated-arc-therapy (VMAT) plans and single-isocenter-multiple-target dynamic-conformal-arc (DCA) plans. Each plan was transferred into Varian Eclipse and recalculated using a linear-Boltzmann-transport-equation based algorithm, Acuros, for comparison. Fine dose grid of 1 mm was used for both algorithms. Target dose was evaluated using dose at a near-maximum reference point (Dref), dose received by 95% of target volume (D95), and mean dose (Dmean). Differences between the two algorithms were compared using these endpoints. Target and plan characteristics such as target volume, planning technique (VMAT vs. DCA), and target-to-isocenter distance were analyzed for potential for correlations with magnitudes of differences in dose.
Results: Among the 443 targets (median=0.5cc, range=0.04-120cc), FMC showed a 2.3±2.6% higher Dref than Acuros. Of them, 53 targets (12%) showed ≥5% Dref differences between algorithms (maximum=15%), in cases including 10 single-target VMAT plans and 15 single-isocenter-multiple-target DCA plans. This suggests that large discrepancies between the two algorithms can occur for either planning technique. D95 and Dmean often showed higher magnitudes of discrepancy than Dref, indicating notable differences in target coverage and mean dose as well. Higher occurrence of large dose differences is associated with smaller target size (odds ratio=15.5, p<10-10, <0.5cc vs. ≥0.5cc). No trend was observed between dose difference and target-to-isocenter distance.
Conclusion: Large calculated target dose between two advanced dose algorithms were observed for brain SRS in some clinical cases. The observed large difference was associated with smaller target size, but independent of planning technique and target-to-isocenter distance.