Cross-Platform Assessment of CBCT-Based Dose Evaluations for Head and Neck Cancer Proton Therapy.

PURPOSE/OBJECTIVE(S): Regularly scheduled repeat CT (reCT) scans are commonly used in proton therapy to evaluate dose and assess need for replanning. To reduce workload and minimize patient procedures, it is desired to move towards on-demand CT acquisition. The purpose of this study is to assess the suitability of CBCT-based dose evaluation to replace prescheduled reCTs for triggering replanning need in head and neck cancer proton therapy. We performed a cross-platform and cross-methodology investigation of three commercially available methods compared to evaluation on reCT.

MATERIALS/METHODS: The study involved 20 patients with oropharyngeal cancer treated at our proton facility in 2020-2022 with weekly reCTs. This study included the first reCT with structure sets transferred from the planning CT. Artificial CTs (aCTs) were manually generated by deforming the planning CT (pCT) to the CBCT (stitched with pCT outside bounding box). All patient data was exported to a treatment planning system, where virtual CTs (vCT) and corrected CBCTs (cCBCTs) were created in a fully automated workflow. Within the treatment planning system, structures were deformably mapped from the pCT to all derived synCTs and clinical treatment plans were recalculated on all images (MC dose engine). For each patient, images were compared using structural similarity index (SSIM) and visually inspected for differences in anatomy by overlaying with the corresponding CBCT. Absolute differences in D99% for high-dose and low-dose CTV, Dmean and D0.03cc for OAR were quantified. We report medians over all patients and methods with (min, max)-range for synCT-reCT and synCT-synCT comparisons.

RESULTS: The highest SSIMs were obtained between the synCTs with a median of 0.978 (0.931 and 0.991) while the synCT-reCT comparisons reached a median of 0.955 (0.901 and 0.976). Visual inspection showed best agreement of cCBCTs followed by vCT and aCT. Pronounced differences in tongue positioning and hyoid bone were found for reCTs. The median D99% difference for high-dose CTV was 0.09 Gy (0.01 Gy and 0.30 Gy) vs. 0.11 Gy (0.00 Gy and 1.24 Gy) and for low-dose CTV was 0.17 Gy (0.01 Gy and 0.35 Gy) vs. 0.54 Gy (0.03 Gy and 1.35 Gy) for synCT-synCT vs. synCT-reCT comparisons. Larger deviations between reCT and synCTs were found for OAR as listed below.

CONCLUSION: We observed better agreement between all synCTs than between synCTs and reCT, despite use of various platforms and methodologies. This is likely due to residual anatomical variations between CBCT and reCT in regions with high mobility. The reCT may therefore not reflect the treatment situation adequately and synCTs be better suited for triggering replanning with on-demand CT scans.