The impact of photon dose calculation algorithms on expected dose distributions in lungs under different respiratory phases.

A planning study was carried out on a cohort of CT datasets from breast patients scanned during different respiratory phases. The aim of the study was to investigate the influence of different air filling in lungs on the calculation accuracy of photon dose algorithms and to identify potential patterns of failure with clinical implications. Selected respiratory phases were free breathing (FB), representative of typical end expiration, and deep inspiration breath hold (DIBH), a typical condition for clinical treatment with respiratory gating. Algorithms investigated were the pencil beam (PBC), the anisotropic analytical algorithm (AAA) and the collapsed cone (CC) from the Varian Eclipse or Philips Pinnacle planning system. Reference benchmark calculations were performed with the Voxel Monte Carlo (VMC++). An analysis was performed in terms of physical quantities inspecting either dose-volume or dose-mass histograms and in terms of an extension to three dimensions of the gamma index of Low. Results were stratified according to a breathing phase and algorithm. Collectives acquired in FB or DIBH showed well-separated average lung density distributions with mean densities of 0.27 +/- 0.04 and 0.16 +/- 0.02 g cm(-3), respectively, and average peak densities of 0.17 +/- 0.03 and 0.09 +/- 0.02 g cm(-3). Analysis of volume-dose or mass-dose histograms proved the expected deviations on PBC results due to the missing lateral transport of electrons with underestimations in the low dose region and overestimations in the high dose region. From the gamma analysis, it resulted that PBC is systematically defective compared to VMC++ over the entire range of lung densities and dose levels with severe violations in both respiratory phases. The fraction of lung voxels with gamma > 1 for PBC reached 25% in DIBH and about 15% in FB. CC and AAA performed, in contrast, similarly and with fractions of lung voxels with gamma > 1 in average inferior to 2% in FB and 4-5% (AAA) or 6-8% (CC) in DIBH. In summary, PBC proved to be severely defective in calculations involving lungs and particularly for cases where specific respiratory phases (e.g. DIBH) are assumed for treatment. In contrast, CC and AAA manifested a high degree of consistency against the Monte Carlo method and provided stable results over the entire range of clinically relevant densities.