Radiotherapy dose calculation in oesophageal cancer: comparison of analytical and Monte Carlo methods

In this work a distributed computing system (RTGrid) has been configured and deployed to provide a statistically robust comparison of Monte Carlo (MC) and analytical dose calculations. 52 clinical oesophageal radiotherapy plans were retrospectively re-calculated using the Pencil Beam Enhanced (PBE) and Collapsed Cone Enhanced (CCE) algorithm within the Oncentra v4.3 radiotherapy (RT) Treatment Planning System (TPS). Simulations were performed using the BEAMnrc and DOSXYZnrc codes. The Computing Environment for Radiotherapy Research (CERR) has been used to calculate Dose Volume Histogram (DVH) parameters such as the volume receiving 95% Dose for the Planning Target Volume (PTV) for the PBE, CCE and MC calculated dose distributions. An initial sample of 12 oesophageal radiotherapy treatment plans were simulated using the RTGrid system. The differences in the DVH parameters between the dose calculation methods, and the variance in the 12 cases, were used to calculate the sample size needed. The required sample size was determined to be 37, so a further 40 oesophageal cases were simulated, following the same method. The median difference in the PTV V95% between CCE and MC in the group of 40 cases was found to be 3%. To choose a suitable test for the statistical significance of the difference, the Shapiro-Wilk test was performed, which showed that the differences between the two sets of PTV V95% values did not follow a Gaussian. Therefore the Wilcoxon matched pairs test was indicated, which showed that the null hypothesis (i.e. that the distributions are the same) was rejected with a p-value less than 0.001, so there is very strong evidence for a difference in the two sets of values of PTV V95%. Similar statistical analyses were performed for other DVH parameters, as well as Conformance Indices used to describe the agreement between the 95% dose and the PTV, and estimates of the Tumour Control Probability (TCP). From the results, the use of MC simulations are recommended when non-soft tissue voxels make up > 60% of the PTV