Precise dosimetric comparison between GAMOS and the collapsed cone convolution algorithm of 4D DOSE accumulated in lung SBRT treatments

Background: It is widely accepted that Monte Carlo dose calculations offers a higher precision that the commercially available dose calculation algorithms. This advantage may be especially relevant for lung Stereotactic Body Radiation Therapy (SBRT), as this is a precise technique applied to an area of big inhomogeneity. Purpose: We conducted a comparative study to reveal the differences between the doses calculated using the Collapsed Cone Convolution algorithm and the GAMOS/Geant4 Monte Carlo calculation for lung cancer patients treated with Stereotactic Body Radiation Therapy on an Elekta Versa HD linac. Methods: For this study a set of ten patient treatments carried out at the Clínica Universidad de Navarra was selected. Theanalysis is based on the comparison of several dosimetric quantities for the Gross Tumor Volume (GTV) and several OrgansAt Risk (OARs), and also a gamma index calculation with distance-to-agreement set to 2 mm and dose difference to 3%, as recommended by ICRU to assess clinical impact. In order to guarantee a small uncertainty in the Monte Carlo calculation of the dosimetric quantities, we studied in detail the validity of different methods that may be used to determine this uncertainty. To compensate for lung movements, a 4D-Cone-beam Computed Tomography (CBCT) was acquired before treatment, whichallowed us to identify eight respiratory phases using a temporal binning. Using commercial MIM software®, we performed a deformable image registration between the eight CT respiration phases to construct the 4D doses. The same procedure was applied for the Treatment Planning System (TPS) dose files and for the Monte Carlo dose files. Results: The differences between the two algorithms reveal the known weaknesses of the Collapsed Cone Convolution (CCC) algorithm for the calculation of lateral doses and in regions of large density change. The comparison between the two algorithms for individual phase doses shows differences up to 5% of the GTV D95 or 3–4 Gy in some OARs, which may have a clinical impact. Nevertheless these differences are reduced for the 4D dose in most quantities under study. Conclusions: Comparing the dose calculated with a Collapsed Cone Convolution algorithm with GAMOS/Geant4 for ten patients and eight respiratory phases, we found some differences that could have a clinical impact. When combining the eight temporal phases into a 4D dose using the MIM Deformable Image Registration software, the differences diminished substantially. Our statistical analysis concludes that dose uncertainty in the voxels with a maximum dose below a given percentage guarantees uncertainty in the dosimetric quantities below that figure