Development of a phantom‐based end‐to‐end QA workflow for online adaptive radiotherapy incorporating high‐resolution 2D detector measurements
Julian Niedling, Robert Schindhelm, Luis Cervantes, Gary Razinskas, Sonja WegenerAbstract
Background
Online adaptive radiotherapy (oART) accounts for anatomical changes by adjusting treatment plans on a daily basis. However, comprehensive plan‐specific quality assurance (QA) is challenging, as pre‐treatment measurements cannot be performed with the patient on the treatment couch. Therefore, an end‐to‐end QA approach, covering imaging, contouring, plan adaptation, dose calculation, and delivery, is essential to ensure accurate and reliable oART.
Purpose
To develop and evaluate an end‐to‐end QA workflow for oART, incorporating high‐resolution detector‐based 2D dose verification and interchangeable phantom inserts.
Methods
A novel QA workflow was implemented using the myQA® SRS phantom with a 2D CMOS detector and a 3D‐printed anthropomorphic pelvis insert. A virtually deformed planning CT was generated to simulate anatomical changes and used as input for the adaptive workflow. The process included CBCT acquisition, manual contour adaptation, plan generation, dose delivery, and comparison of measured and calculated dose distributions via Gamma Analysis. Workflow feasibility was tested through repeated measurements. Error detectability was evaluated by altering gantry/collimator angles, monitor units (MU), phantom positioning, and CBCT presets. Confidence limits, calculated following AAPM TG‐119 methodology, were used as plan‐specific thresholds for evaluating Gamma Pass Rates (GPR). Corresponding dose‐volume histogram (DVH) parameters, using D98% for the planning target volume (PTV) and V95% for organs at risk (OARs), were calculated.
Results
Ethos auto‐contouring did not perform reliably on the phantom geometry, necessitating manual contouring. All workflow repetitions produced high GPR values for 3 %/2 mm, 2 %/2 mm, 3 %/1 mm, and 2 %/1 mm global criteria between measured and calculated dose. Confidence limits served as effective thresholds for identifying plan deviations. Introduced errors, including gantry angle deviations > 3° and collimator angle deviations ≥ 3°, MU variations ≥ 3 %, incorrect CBCT presets, and 3 mm setup shifts, were clearly detectable across all criteria. Smaller deviations, such as 1° angular changes or 1 % MU variation, resulting in minor dosimetric effects residing at or below 1.2 % for the PTV and 5.6 % for OARs, yielded GPR values above the Confidence limit.
Conclusions
The proposed workflow is feasible, reproducible, and capable of reliably identifying plan‐specific deviations in Ethos oART. High‐resolution 2D measurements and dose comparison provided robust verification of adaptive dose calculation and delivery. While limited by manual contouring and data‐handling constraints inherent to the Ethos platform, this method represents a practical end‐to‐end QA approach for clinical oART implementation.