Calculation algorithms and penumbra: Underestimation of dose in organs at risk in dosimetry audits.

PURPOSE

The aim of this study is to investigate overdose to organs at risk (OARs) observed in dosimetry audits in Monte Carlo (MC) algorithms and Linear Boltzmann Transport Equation (LBTE) algorithms. The impact of penumbra modeling on OAR dose was assessed with the adjustment of MC modeling parameters and the clinical relevance of the audit cases was explored with a planning study of spine and head and neck (H&N) patient cases.

METHODS

Dosimetric audits performed by the Australian Clinical Dosimetry Service (ACDS) of 43 anthropomorphic spine plans and 1318 C-shaped target plans compared the planned dose to doses measured with ion chamber, microdiamond, film, and ion chamber array. An MC EGSnrc model was created to simulate the C-shape target case. The electron cut-off energy E was set at 500, 200, and 10 keV, and differences between 1 and 3 mm voxel were calculated. A planning study with 10 patient stereotactic body radiotherapy (SBRT) spine plans and 10 patient H&N plans was calculated in both Acuros XB (AXB) v15.6.06 and Anisotropic Analytical Algorithm (AAA) v15.6.06. The patient contour was overridden to water as only the penumbral differences between the two different algorithms were under investigation.

RESULTS

The dosimetry audit results show that for the SBRT spine case, plans calculated in AXB are colder than what is measured in the spinal cord by 5%-10%. This was also observed for other audit cases where a C-shape target is wrapped around an OAR where the plans were colder by 3%-10%. Plans calculated with Monaco MC were colder than measurements by approximately 7% with the OAR surround by a C-shape target, but these differences were not noted in the SBRT spine case. Results from the clinical patient plans showed that the AXB was on average 7.4% colder than AAA when comparing the minimum dose in the spinal cord OAR. This average difference between AXB and AAA reduced to 4.5% when using the more clinically relevant metric of maximum dose in the spinal cord. For the H&N plans, AXB was cooler on average than AAA in the spinal cord OAR (1.1%), left parotid (1.7%), and right parotid (2.3%). The EGSnrc investigation also noted similar, but smaller differences. The beam penumbra modeled by E  = 500 keV was steeper than the beam penumbra modeled by E  = 10 keV as the full scatter is not accounted for, which resulted in less dose being calculated in a central OAR region where the penumbra contributes much of the dose. The dose difference when using 2.5 mm voxels of the center of the OAR between 500 and 10 keV was 3%, reducing to 1% between 200 and 10 keV.

CONCLUSIONS

Lack of full penumbral modeling due to approximations in the algorithms in MC based or LBTE algorithms are a contributing factor as to why these algorithms under-predict the dose to OAR when the treatment volume is wrapped around the OAR. The penumbra modeling approximations also contribute to AXB plans predicting colder doses than AAA in areas that are in the vicinity of beam penumbra. This effect is magnified in regions where there are many beam penumbras, for example in the spinal cord for spine SBRT cases.