Implementation of Acuros XB Dose Calculation in to Clinical Radiation Therapy Workflows

Introduction: Stereotactic body radiation therapy (SBRT) is a common treatment techniquethat can be used to treat tumors for multiple cancer sites. Density heterogeneity in the target volume and beam path combined with small treatment fields has made dose calculation in lung SBRT difficult. Dose calculation algorithms used historically have difficulty modelling the extreme density heterogeneity present in lung SBRT and have been shown to overestimate the dose delivered to tumors situated in the lung parenchyma. Recently, more advanced algorithms that directly model heterogeneity have been implemented for clinical treatment planning. The limited accuracy of historically utilized dose calculation algorithms has raised questions about their effects on local control due to the possibility of tumor underdosing. The first part of this work establishes a proper dose normalization technique when implementing these advanced algorithms for treatment planning in order to keep consistent radiation beam settings and to quantify the dosimetric effect of various dose normalizations. The second aim is to quantify the effects dosimetric accuracy has on local control in lung SBRT.

Materials/Methods: 87 lung SBRT plans with doses originally calculated with the AnisotropicAnalytical Algorithm (AAA) had their doses recalculated with the new Acuros XB (AXB) algorithm, which is able to directly model the heterogeneity of the lungs and treatment volume. After recalculation, the plan was normalized to the planning target volume (PTV) D95%, internal target volume (ITV) D99%, and to match the original PTV coverage. The percentage change in total monitor units (MU) between the AXB renormalized plans and the original AAA plans were calculated to quantify how the delivered radiation would change when implementing the AXB algorithm for treatment planning. Percentage changes in relevant PTV and ITV dose metrics as well as absolute changes in relevant organ at risk. (OAR) dose metrics were quantified to compare plan dosimetry. OAR doses were also compared to the current institutional planning objectives to investigate the feasibility of meeting the current objectives with the new algorithm. 162 patients previously treated with SBRT were selected from a retrospective protocol comparing the efficacy of SBRT and surgery for treatment of early-stage non-small cell lung cancer. Plans had their doses originally computed with the Pencil Beam Convolution (PBC, n = 8) algorithm or AAA (n = 156). Each plan was recalculated with AXB with identical beam settings. A subset was also recalculated with Monte Carlo to validate the accuracy of the AXB calculations. Percentage changes in relevant PTV and ITV biologically effective doses (BED) were calculated between the original and AXB plans to quantify the magnitude of the dosimetric differences between the old and new algorithm. A multivariable linear regression was performed to investigate which patient and treatment parameters influenced the magnitude of these dosimetric changes. A competing risk analysis was performed to quantify the association between the magnitude of the dosimetric changes and local failure.

Results: Normalizing the AXB plan to the PTV D95% and keeping the original PTVcoverage typically resulted in a total MU increase (average increase of 7.0% and 7.9%, respectively) while normalizing to the ITV D99% resulted in similar total MU (average increase of 0.31%). When normalizing to the PTV D95%, the AXB plans had increased PTV and ITV D1%[Gy] (median increases of 3.4% and 3.2%, respectively) while normalizing to the ITV D99% showed a median 1.9% decrease. Normalizing the AXB plans to the PTV D95% typically resulted in increased OAR dose for all OARs and an inferior ability to meet the OAR planning constraints. Reoptimization of the renormalized plans showed the current OAR objectives to be manageable when using the AXB algorithm. The AXB dose calculations were much more consistent with Monte Carlo than were the original dose calculations. A large range of dosimetric decreases upon recalculation with AXB were observed for both patients who failed locally and those who were controlled. Higher beam energy was found to increase the magnitude of the dosimetric decreases (expected decrease in PTV mean BED of 3.6%, 5.9%, and 9.1% when using 6X, 10X, or 15X, respectively). Total lung volume was also associated with an increased magnitude of dosimetric decrease (expected decerease of 0.8% per 500 cc for the PTV mean BED). The median follow-up time of the cohort was 26 months. 15 patients experienced local failures. Upon univariate analysis, the dosimetric decreases in the PTV and ITV D1% BED were found to be associated with local failure (hazard ratio (HR) of 0.89 (p=0.04) and 0.87 (p=0.02), respectively). Upon multivariate analysis, the dosimetric decrease in the ITV D1% BED remained significant when controlling for PTV volume (HR=0.89 (p=0.04)).

Conclusions: More accurate dose calculation algorithms are beginning to be implementedfor clinical treatment planning. When implementing these new algorithms, issues arise with dose normalization due to the potential for vast differences between the dose distributions calculated with the different algorithms. Normalizing the dose to the PTV D95% in the AXB plan will result in a delivered dose increase relative to a AAA plan while normalizing to the ITV D99% will keep similar delivered doses between the plans. Dose metrics typically increase when normalizing to the PTV D95% (for targets and OARs) while normalizing to the ITV D99% typically decreased the reported dose metrics. The OAR planning objectives are manageable using the AXB algorithm. Many factors are related to the magnitude of the dosimetric decreases observed when recalculating plans with AXB, including but not limited to beam energy and lung volume. Most of the investigated dose metrics were not associated with local failure, but the change in the PTV and ITV D1% BEDs were found to be associated with local failure in the univariate analysis.