Purpose: This work presents the improvement in lung SBRT internal target volumes generated with an automatic algorithm that used voxelized position uncertainty models derived from individual 4DCT simulation scans.
Methods: For 19 previously treated lung SBRT targets, deformable image registration was applied to all respiratory phases of 4DCT simulation scans. From the resulting deformation vector fields, voxelized models of gross tumor volume motion were created and fed into an in-house, target-generating algorithm to produce a new internal target volume, the aITV. The efficacy of the aITV was evaluated by comparing its volume and the coverage it provided to each gross tumor volume voxel with those of the clinically-utilized ITV. In addition, the magnitude of an isotropic margin that could be added to the aITV and yield a volume equivalent to the clinical ITV was determined.
Results: The target-generating algorithm was used to create aITVs that covered at least 95% of the tumor voxels at least 95% of the time according to the voxelized motion model. The aITV increased the coverage of the 5th-percentile-most-covered voxel 50.2% (range -4.9% to 95%) (p<0.001) with a volume that was 29.6% less than the clinical ITV (range 11.4% to 57.3%) (p<0.001). Furthermore, an extra 1.8 mm isotropic margin (range 0.5 to 3.2 mm) was able to be added to the aITV and still not yield a volume greater than the clinical ITV.
Conclusion: An automatic, uncertainty model-guided, target-generating algorithm was used to create lung SBRT internal target volumes that, when compared to the clinically-utilized ITV, provided both a significant increase in tumor voxel coverage and a significant decrease in volume. The greater efficacy of this target has the potential to increase the therapeutic ratio for lung SBRT treatments, and the algorithm can be extended to other radiotherapy treatment sites exhibiting positional uncertainties.