Influence of dose calculation algorithms on isotoxic dose-escalation of non-small cell lung cancer radiotherapy

Background and purpose: A series of phase I/II clinical trials are being initiated in several UK centres to explore the use of dose-escalated schedules for the treatment of non-small cell lung cancer (NSCLC). Among them the IDEAL-CRT trial (ISRCTN12155469) will investigate the introduction of individualised 'isotoxic' treatment schedules based on the relative mean lung normalised total dose (rNTDmean), an estimator related to lung toxicity. Since treatment planning will be performed using different treatment planning systems (TPSs), for the quality assurance of the trial we have carried out work to quantify the influence of dose calculation algorithms based on the determination of rNTDmean and on the choice of individualised prescription doses. Material and methods: Twenty-five patient plans with stage I, II and III NSCLC were calculated, with the same prescription dose, using the Adaptive Convolve (AC) and Collapsed Cone (CC) algorithms of the Pinnacle TPS, the pencil beam convolution (PBC) and AAA algorithms of Eclipse, and the CC and pencil beam (PB) algorithms of Oncentra Masterplan (OMP). For the paired-lungs-GTV structure, dose-volume histograms were obtained and used to calculate the corresponding rNTDmean values and results obtained with the different algorithms were compared. Results: For most (19 out of 25) of the patients studied, no algorithm-to-algorithm differences were seen in dose prescription based on rNTDmean. For the other 6 patients differences were within 2.3 Gy, except in one case where the difference was 4 Gy. Conclusions: For the IDEAL-CRT trial no corrections need to be applied to the value of rNTDmean calculated using any of the more advanced convolution/superposition algorithms studied in this work. For the two pencil beam algorithms analysed, no correction is necessary for the data obtained with the Eclipse-PBC, while for OMP-PB data a small correction needs to be applied, by using a scaling factor, to make prescription doses consistent with the other algorithms investigated.