Author: Luis E. Fong de los Santos, Douglas John Moseley, Satomi Shiraishi, Cenji Yu 👨🔬
Affiliation: Mayo Clinic 🌍
Purpose: To quantitatively detect errors in vertebral body localization using normalized cross correlation (NCC)
Methods: Volumetric images of a thoracic end-to-end SBRT phantom with a removable spine were acquired for simulation and treatment purposes. CT scans were acquired using the Siemens SOMATOM Go Simulator under the thoracic protocol. CBCT images were obtained using thorax, pelvis, head, and spotlight protocols on both the Truebeam and Edge Hypersight imagers. Rigid registrations between the planning CT and CBCT images were performed on the treatment console to establish a baseline for optimal vertebral alignment. To simulate vertebral body shifts encountered during clinical treatment, phantom CBCTs were repositioned in superior and inferior directions in 1mm increment, with a total shift range of 2cm in each direction. Registration tracking volumes (RTV) measuring 8cm x 8cm x 4cm were defined around the spinal cord at the targeted vertebral level to evaluate similarities between the planning CT and CBCT images. Normalized cross correlations (NCC) were calculated within the RTVs to compare the planning CTs and CBCTs across all eight imaging protocols.
Results: The NCC values ranged from 0.799±0.010 to 0.993±0.002. With a 2mm vertebral body shift, the NCC decreased to 0.943±0.005 in the superior direction and 0.956±0.004 in the inferior direction. As the shift increased, the NCC continued to decrease, while it improved as the targeted vertebral body approached the adjacent vertebral body. The maximum NCC achieved, apart from the optimal match location, was 0.901±0.006. The largest NCC standard deviation across imaging protocols at a CBCT position was 0.011. The difference between the worst NCC at optimal match position and the highest NCC at the adjacent vertebral body position was 0.083.
Conclusion: The NCC successfully detected vertebral body shifts across various CBCT imaging systems and protocols.