Two papers on RadCalc’s exceptional accuracy
RadCalc’s exceptional accuracy provides confidence and reliability for all field sizes in a single model with both the Collapsed Cone Superposition Convolution and BEAMnrc Monte Carlo.
Evaluation of the RadCalc collapsed cone dose calculation algorithm against measured data.
Authors: Richmond N, Chester K, Manley S. Med Dosim
2023 May 8:S0958-3947(23)00042-0. doi: 10.1016/j.meddos.2023.04.004. Epub ahead of print. PMID: 37164787.
Read publication
The accuracy of the RadCalc 3D Collapsed Cone algorithm for all field sizes was validated directly with measurements for both simple and complex geometries, with and without heterogeneities, using international guidelines and published criteria such as TG 114 and TG 219.
The modelling was performed with the user’s custom beam data and with independent validation of the machine characteristics, including the Radiation Light Field Offset. The calculations were performed with both NVIDIA Tesla K20 and RTX 3080 GPUs, demonstrating RadCalc’s hardware flexibility.
RadCalc’s Collapsed Cone algorithm from version 7.1.4.1 was evaluated, as well as the changes implemented in version 7.2.2.0. The newest improvements in version 7.2.3.1 were not evaluated as part of this work. However, it is not surprising to see that even with version 7.2.2.0, the complex IMRT and VMAT deliveries met the recommendations of TG 219 with IMRT fields using a 2% and 2mm criteria and for the most complex nasopharynx VMAT plan meeting 3%/2mm criteria. Overall, for the modulated fields there was especially good agreement of less than 1mm DTA in the areas of steep dose gradients.
Open fields had PDD comparisons of less than 0.5% differences, and Off Axis Ratios (OAR) within 2% in the central 80%. Additionally, the researchers performed dose comparisons for heterogeneous phantoms, including a 2cm stepped phantom on top of the water surface for an enface and oblique beam geometry, as well as utilizing Lung, Bone, Air and Mediastinum geometries, of which the results were generally within 3.5% of the measured dose. These results are better than the 2.5% dose differences recommended for simple open or MLC shaped static fields in homogenous medium and the action limits that are recommended when doses exceed 5% difference for heterogeneous calculations (TG 114).
Tuning and validation of the new RadCalc 3D Monte Carlo based pre-treatment plan verification tool.
Authors: Sceni et al.
Journal of Mechanics in Medicine and Biology. doi: 10.1142/S021951942340047X
The results of RadCalc’s 3D Monte Carlo algorithm secondary check on the patient’s heterogeneous CT datasets were compared against on-couch homogeneous phantom measurements after fine tuning the models in RadCalc with a Gamma criteria of 2%/2mm and low dose thresholds of 50%. 70 VMAT plans were used for clinical validation of RadCalc’s 3D Monte Carlo algorithm for 6x, 10x, 6FFF and 10FFF against Eclipse v13.7 for both AAA and Acuros XB. Of the 70 plans, 20 were used for tuning and the other 50 were utilized as a validation set.
The RadCalc MC modeling process allows the user to choose the spot size and mean energy that best fits 3 open fields. From this spot size and mean energy combination, a BEAMnrc modeled machine is loaded, with every physical component modeled. This unique auto modeling method provides near-instantaneous beams with only one parameter that needs to be fine-tuned: the additional Radiation Light Field Offset (ARLF), better known in the Varian world as the Dosimetric Leaf Gap (DLG).
The authors quote their DLG for each energy, as well as the resulting ARLF from the model tuning performed. The authors demonstrate the accuracy of RadCalc’s Monte Carlo against the Eclipse algorithms and the on-couch homogenous phantom measurements against the Eclipse algorithms as well. As is the topic of the up-and-coming TG 360, the authors performed statistical methods on the comparison of the gamma passing rates and utilized ROC curve analysis to set the acceptable plans for the on-couch measurements and the RadCalc Monte Carlo calculations as the 95th and 90th percentile respectively. The confusion matrix, including the number of True Positives/Negatives and False Positives/Negatives is demonstrated, as well as Gamma Passing Rate comparisons against AAA/Acuros XB, and a box chart that also includes the on-couch measurements. In summary, these data show a high degree of agreement between the RadCalc MC and Acuros XB calculations, especially for the lung subset used.
As the authors conclude, after tuning, the RadCalc’s 3D Monte Carlo Algorithm provides a solution to independently verify treatment plans directly on the patient’s CT with sensitivities and specificities similar to on-couch phantom solutions, with the added benefit of detecting inaccuracies in tissue inhomogeneities that homogeneous on-couch phantoms are unable to detect.