CBCT-Based Dose Monitoring and Adaptive Planning Triggers in Head and Neck PBS Proton Therapy

SIMPLE SUMMARY: Head and neck cancer patients require adaptive radiation therapy due to aggressive tumor response, changes in anatomy, and difficulties with setup reproducibility. Additionally, protons are more sensitive to patient setup uncertainty and anatomical change than photons. At our institution, verification CTs are often utilized to evaluate changes in anatomy and dose coverage during the treatment course. Nevertheless, they are limited by low frequency and an inability to detect changes in treatment setup. Furthermore, our current clinical workflow does not have definitive thresholds for adaptive therapy plan review, meaning every verification CT must be reviewed by a physician and physicist. We analyzed the feasibility of higher-frequency dose monitoring and triggering a plan review for adaptive proton therapy by recalculating the plan dose on synthetic CTs produced from daily patient setup cone-beam CTs. Plan review triggers were established following the analysis of cohort data. Two synthetic CT algorithms were developed and evaluated. Both resulted in images that outperformed the verification CTs in dosimetric accuracy. A more efficient and accurate adaptive proton therapy workflow was successfully established. ABSTRACT: Purpose: To investigate the feasibility of using cone-beam computed tomography (CBCT)-derived synthetic CTs to monitor the daily dose and trigger a plan review for adaptive proton therapy (APT) in head and neck cancer (HNC) patients. Methods: For 84 HNC patients treated with proton pencil-beam scanning (PBS), same-day CBCT and verification CT (vfCT) pairs were retrospectively collected. The ground truth CT (gtCT) was created by deforming the vfCT to the same-day CBCT, and it was then used as a dosimetric baseline and for establishing plan review trigger recommendations. Two different synthetic CT algorithms were tested; the corrected CBCT (corrCBCT) was created using an iterative image correction method and the virtual CT (virtCT) was created by deforming the planning CT to the CBCT, followed by a low-density masking process. Clinical treatment plans were recalculated on the image sets for evaluation. Results: Plan review trigger criteria for adaptive therapy were established after closely reviewing the cohort data. Compared to the vfCT, the corrCBCT and virtCT reliably produced dosimetric data more similar to the gtCT. The average discrepancy in D99 for high-risk clinical target volumes (CTV) was 1.1%, 0.7%, and 0.4% and for standard-risk CTVs was 1.8%, 0.5%, and 0.5% for the vfCT, corrCBCT, and virtCT, respectively. Conclusion: Streamlined APT has been achieved with the proposed plan review criteria and CBCT-based synthetic CT workflow.