At the present, treatment planning systems (TPS) used in radiotherapy treatments use determinist correlations based on measurements in water to evaluate doses in the volume of interest and dose distributions around it. Nevertheless, it is well known that doses assigned with this type of planner can be problematic, especially in the presence of heterogeneities.
The present work has developed a computational model using the Monte Carlo (MC) code MCNP5 (Monte Carlo N-Particle) for the simulation of a 6 MeV photon beam emitted by Elekta Precise medical linear accelerator treatment head. The model includes the major components of the accelerator head and the cube-shaped heterogeneous water tank ''RFA-300''.
A low-density heterogeneity has been placed inside this water tank. It consists of a extruded polystyrene piece (97% air and 3% polystyrene) whose dimensions are 30 cm  10 cm  8 cm and with a density of 0.0311 g/cm 3 .
Calculations were performed for a photon beam setting 10 cm  10 cm and 20 cm  20 cm irradiation field sizes at 100 cm distance from source. The MC simulation is able to predict the absorbed dose distribution within the water tank using the *F8 or FMESH4 tally. These results have been compared with experimental values measured at the Hospital Clı ´nic Universitari de Valencia.
Dosimetric parameters calculated by simulation at the water tank and the experimental measures agreed, with an average deviation inside the heterogeneity region of 3%.
Simulation results have been also compared with dose curves predicted by a commercial TPS in the same irradiation conditions, focusing attention on the accuracy that both systems reach in the dose calculation at the interphase zone and inside the heterogeneity. In contrast, TPS results overestimate the dose inside the heterogeneity and after it, with a maximum deviation of 7% for the 6 MeV photon beam and a field size of 20 cm  20 cm.
We can conclude that the algorithms of computation of the TPS are not able to predict the variation of dose in the heterogeneous zones with the same accuracy as MC methods, since MCNP5 provides more trustworthy results and fits the experimental values better in the heterogeneous region.