Reply to Comment on 'egs_brachy: a versatile and fast Monte Carlo code for brachytherapy'

Rowan M Thomson, Randle E P Taylor, Marc J P Chamberland, D W O Rogers
2018 Physics in Medicine and Biology  
We thank Dr Yegin for his comments on our article about egs_brachy (Chamberland et al 2016) since it allows us to deal with the mistake he pointed out and to provide more of the benchmarking that he felt was needed. In his comment on our article, Dr Yegin argues that the multiseed simulation results in section 3.1 and associated fits to data (figure 4) are wrong, and that it has not been demonstrated that egs_brachy correctly simulates multiseed configurations. Figure 4 of our paper presents
more » ... normalized frequency distribution of dose differences between egs_brachy and BrachyDose (in units of the combined statistical uncertainty, i.e. the quadrature sum of the statistical uncertainty on dose in each voxel) for two multiseed configurations, a prostate LDR implant consisting of 100 125 I model 6711 seeds (figure 4(a)) and a 16 mm diameter COMS eye plaque containing 13 125 I model 6711 seeds (figure 4(b)). We confirm that figure 4's data (black histograms) representing the dose differences are correct. However, there was an error in carrying out the fit due to a typographical error in copying out the equation to fit to the data (blue smooth curves in figures 4(a) and (b)), and the α and Δ parameters indicated in our article are incorrect. Fixing this transcription error in the fitting function means we no longer achieve a stable fit with the Kawrakow-Fippel approach (Kawrakow and Fippel 2000) . This approach aims to separate systematic discrepancies from statistical uncertainties, however, achieving a fit with this approach was not needed for benchmarking, i.e. the fit is not necessary since one can work directly with the underlying data. The presentation of data in terms of the difference in egs_brachy and BrachyDose results in units of the combined statistical uncertainty (figure 4 in our original article) emphasizes the differences in dose distributions, and demonstrates that there are systematic differences between the codes. However, as noted in our article, these systematic discrepancies are small at 0.2% (on average) for the prostate scenario (original figure 4(a) ) and 0.5% (on average) for the eye plaque case (original figure 4(b) ). For the eye plaque scenario, figure 1 here provides an alternate presentation of the data in terms of 'local' and 'global' dose difference ratios as defined by Ma et al (2017), ∆D LOCAL = (D BD (r) − D eb (r))/D eb (r) and ∆D GLOBAL = (D BD (r) − D eb (r))/D eb (r ref ) where D BD and D eb denote BrachyDose and egs_brachy doses, respectively, and the reference point, r ref , is the tumour apex at a height of 5 mm from the inner sclera along the plaque's central axis. As these figures demonstrate, and as argued in our original paper, there are discrepancies in the dose distributions produced by the two codes, however, they are relatively small in magnitude given that the average combined uncertainty of the comparisons is 0.6%. Members of the joint AAPM/ESTRO/ABG Working Group on model-based dose calculation algorithms in brachytherapy (Ma et al 2017) reported similar discrepancies in comparisons between diverse brachytherapy Monte Carlo codes with the conclusion that 'All MC codes gave equivalent results within statistical uncertainties'. Dr Yegin argues that, because egs_brachy and BrachyDose are both EGSnrc-based codes, dose distributions produced by egs_brachy and BrachyDose should be in agreement within statistical uncertainties as long as the simulation geometries and transport parameters are the same. We note that, although egs_brachy and BrachyDose both use EGSnrc for radiation transport, they are distinct codes, each having different scoring routines, different approaches for modelling geometries and sources, and so on. A simulation by either code involves Abstract We respond to the comments by Dr Yegin by identifying the source of an error in a fit in our original paper but arguing that the lack of a fit does not affect the conclusion based on the raw data that egs_brachy is an accurate code and we provide further benchmarking data to demonstrate this point. REPLY RECEIVED
doi:10.1088/1361-6560/aa9ea2 pmid:29192606 fatcat:okjwshbjrvex5p3idubtitdjge