Purpose: This article presents an improved pencil-beam dose calculation formalism based on an experimental kernel obtained by deconvolution. The new algorithm makes it possible to calculate the absorbed dose for all field sizes.
Methods: The authors have enhanced their previous work [J. D. Azcona and J. Burguete, Med. Phys. 35, 248-259 (2008)] by correcting the kernel tail representing the contribution to the absorbed dose far from the photon interaction point. The correction was performed by comparing the calculated and measured output factors. Dose distributions and absolute dose values calculated using the new formalism have been compared to measurements. The agreement between calculated and measured dose distributions was evaluated according to the gamma-index criteria. In addition, 35 individual intensity-modulated radiation therapy (IMRT) fields were calculated and measured in polystyrene using an ionization chamber. Furthermore, a series of 541 IMRT fields was calculated using the algorithm proposed here and using a commercial IMRT optimization and calculation software package. Comparisons were made between the calculations at single points located at the isocenter for all the beams, as well as between beams grouped by anatomic location.
Results: The percentage of points passing the gamma-index criteria (3%, 3 mm) when comparing calculated and measured dose distributions is generally greater than 99% for the cases studied. The agreement between the calculations and the experimental measurements generally lies in the +/- 2% interval for single points, with a mean value of 0.2%. The agreement between calculations using the proposed algorithm and using a commercial treatment planning system is also between +/- 5%.
Conclusions: An improved algorithm based on an experimental pencil-beam kernel obtained by deconvolution has been developed. It has been validated clinically and promises to be a valuable tool for IMRT quality assurance as an independent calculation system for monitor units and dose distributions. An important point is that the algorithm presented here uses an experimental kernel, which is therefore independent of Monte-Carlo-calculated kernels.