Comparison of CT number calibration techniques for CBCT-based dose calculation

PURPOSE: The aim of this work was to compare and validate various computed tomography (CT) number calibration techniques with respect to cone beam CT (CBCT) dose calculation accuracy. METHODS: CBCT dose calculation accuracy was assessed for pelvic, lung, and head and neck (H&N) treatment sites for two approaches: (1) physics-based scatter correction methods (CBCT(r)); (2) density override approaches including assigning water density to the entire CBCT (W), assignment of either water or bone density (WB), and assignment of either water or lung density (WL). Methods for CBCT density assignment within a commercially available treatment planning system (RS(auto)), where CBCT voxels are binned into six density levels, were assessed and validated. Dose-difference maps and dose-volume statistics were used to compare the CBCT dose distributions with the ground truth of a planning CT acquired the same day as the CBCT. RESULTS: For pelvic cases, all CTN calibration methods resulted in average dose-volume deviations below 1.5 %. RS(auto) provided larger than average errors for pelvic treatments for patients with large amounts of adipose tissue. For H&N cases, all CTN calibration methods resulted in average dose-volume differences below 1.0 % with CBCT(r) (0.5 %) and RS(auto) (0.6 %) performing best. For lung cases, WL and RS(auto) methods generated dose distributions most similar to the ground truth. CONCLUSION: The RS(auto) density override approach is an attractive option for CTN adjustments for a variety of anatomical sites. RS(auto) methods were validated, resulting in dose calculations that were consistent with those calculated on diagnostic-quality CT images, for CBCT images acquired of the lung, for patients receiving pelvic RT in cases without excess adipose tissue, and for H&N cases.