Analysis of the bias induced by voxel and unstructured mesh Monte Carlo models for the MCNP6 code in orthovoltage applications

The use of Monte Carlo methods for the set up of Treatment Planning Systems (TPS) in radiotherapy applications is a current standard. The most advanced modeling techniques aim at directly link the output of CT scans to a patient-specific model build up instead of standard phantoms and look-up tables. This link represents a critical step since even the most accurate segmentation and organ volume definition must be translated into a suitable input for the Monte Carlo code. During that step, the segmented volume is usually mapped on a regular geometrical lattice (voxels). A more sophisticated option appears to be a volume description based on an Unstructured Mesh (UM) geometry typical of current finite element codes. In this paper, we compared the two approaches analyzing the bias induced by the different choices. Starting from anonymous patient DICOM files coming from a CT scan, suitable segmentation and volume definition have been carried out and voxel and UM-based equivalent models for the Monte Carlo code MCNP6 have been built. Some computational phantoms, covering some significant portion of the human body (head and lower limb), have been used as a benchmark of the dose distribution obtained from X-ray sources commonly used in orthovoltage applications. The specific clinical application has been chosen because, despite being used on a growing number of patients and with multi-Gy doses, treatment planning is still based on poorly characterized dose mapping systems such as look-up tables or low-resolution computational phantoms for MC codes. Also, experimental measurements on phantom slabs irradiated by an X-ray source were carried out preliminarily to validate the simulated radiation source. As shown in the results, the UM computational phantoms (built through the Simpleware SCAN-IP™ tool) can reduce the bias induced by the regularity of the classical cubic voxel geometry, give a more accurate description of volumes and complex surfaces and, thanks to an optimized discretization of the volumes, are also able to reduce the computational work. For the same UM models, various comparisons with different voxel sizes have been produced to investigate the dose distributions and evaluate the voxelization effects. The comparison shows a convergence pattern of the voxel model to the UM one. It has been possible to verify that the most significant bias occurs where the dose gradient is higher, along the beam borders and at tissue interfaces. The simulations showed that the UM can be really effective and reliable to compute the dose distributions within computational anthropomorphic phantoms obtained from the patient’s CT scans.