Comparison of Monte Carlo with pencil beam dosimetry for lung CyberKnife SBRT, correlation with local recurrence

Object Promising results have been obtained using stereotactic body radiation therapy (SBRT) for early-stage lung cancer. The calculation of dose in pulmonary parenchyma can be inaccurate. Methods We retrospectively analyzed 47 cases treated over a 2-year period with CyberKnife SBRT, planned with the standard pencil beam (PB) algorithm. Cases were a mixture of early-stage lung cancer and oligometastatic cases. The median prescribed dose was 50 Gy, in four or five fractions. We compared the planned dose with the dose actually delivered, as estimated with Monte Carlo (MC) dosimetry to the 1 % level. We correlated the dosimetric deficiencies with recurrences, using deformable registration to determine the dose delivered to the site of recurrence. Results With a median follow-up of 2 years, the local control at 1 year was 90 %, declining to 70 % at 2 years. The total number of local recurrences was 10, and 8 of these died with progressive disease. Two recurrences occurred synchronously with metastases, and 2 recurrences were in palliative cases treated to lesser doses with tight margins, and disease was never cleared locally. Monte Carlo calculations showed that the mean dose delivered to the planning target volume (PTV), averaged over all cases, was 7 % lower than planned. Most cases were planned with an expansion on the PTV (PTVmicro= GTV+ 8 mm expansion in the lung+ 3 mm) representing a region at risk for microscopic extension and intended to receive a minimum dose of 80 % of the prescription dose. Monte Carlo calculations showed that the minimum dose to this structure, averaged over all cases, was 47% lower than the intended dose. For cases that recurred, the mean dose to the PTVmicro was 8 % lower than intended, while only 2 % low for those controlled. There were no other significant differences in target coverage between patients with local control and local recurrence. The PB algorithm and MC estimates for pulmonary exposure were assessed, recording the V5, V10, and V20 for the ipsilateral and total lung volumes. These estimates roughly agreed for the two algorithms, with the MC results almost universally lower than PB, lower by an absolute 1-3 % on average. Conclusions Without the use of MC planning, target structures were substantially underdosed. Local failures were associated with PTVmicro undercoverage, which suggests delivering a therapeutic dose to this expanded microscopic disease target volume is beneficial. MC dosimetry is preferable for lung SBRT, while the PB algorithm was adequate for predicting pulmonary toxicity.