Four-dimensional Monte Carlo investigation of organ motion in radiotherapy for lung cancer

A limitation of current dose calculation algorithms employed in radiotherapy treatment planning is the assumption that the patient's anatomy is static throughout the imaging, planning and delivery. 4D dose calculation methods employ non-linear image registration to determine the cumulative dose received in a deforming anatomy. In this work, we developed a 4D Monte Carlo dose calculation code, designated defDOSXYZ, which determines the dose received in a deforming voxel grid. Voxel deformations were determined from deformation vectors resulting from non-linear image registration between images of the reference and target states. The ANIMAL non-linear image registration algorithm was implemented for registration of thoracic 4D CT images. Modifications were performed to ANIMAL to minimize deformation vector discontinuities. A method for correcting artifacts in 4D CT images was developed which uses non-linear image registration to interpolate voxel intensities from temporally adjacent artifact-free images. Dose calculations in deforming phantoms and 4D CT patient data using defDOSXYZ were compared to conventional center-of-mass (COM) and trilinear (TL) dose remapping methods.;The accuracy of non-linear image registration between inhale and exhale images for 5 lung patients was found to be within 2 mm which was deemed acceptable for clinical dose calculations. Temporal interpolation using ANIMAL was demonstrated to improve image quality in 4D data sets containing motion artifacts.;Comparison of dose remapping from Inhale to Exhale in an anatomical breathing phantom revealed that interpolation methods underestimate the dose in the penumbra and near the surface. defDOSXYZ calculations were also compared with two dose remapping methods in 4D CT patient data. Systematic offsets between the dose calculation methods were noted which were attributed to inconsistent handling of voxel mass conservation in the image registration and dose calculations. A mass-consistent comparison of defDOSXYZ calculations and remapping calculations for clinically relevant planning scenarios and dose grid sizes revealed discrepancies in regions of steep gradients which was consistent with the phantom studies. No clinically significant differences in planning volume doses were noted between all three dose calculation methods, although conventional dose remapping failed to predict certain details of the cumulative dose distribution which may be important for 4D conformal treatment planning.;defDOSXYZ calculations were determined to be accurate to within 1% by comparison with DOSXYZ calculations and internal consistency checks. Conventional dose remapping methods were found to underestimate the dose by 29% and 8%, on average, when remapping dose from Exhale to Inhale within simple deforming phantoms with voxel sizes of 1 cm and 0.5 cm, respectively. These discrepancies were reduced to 0.2% for voxel sizes of 0.25 cm and smaller, however dose errors of 20-30% still existed in regions of steep dose gradients.