3085 - AAPM Guideline-Based Evaluation of Automated Dose Reconstruction using Cone Beam CT for Offline Adaptive Lung Radiation Therapy
Presenter(s)
A. Mylonas1, N. Hardcastle1, S. Kumar1, R. Finnegan1, B. Whelan1, D. T. Nguyen1, M. Jameson2,3, and L. Mejnertsen1; 1SeeTreat Pty Ltd, Sydney, NSW, Australia, 2GenesisCare, Sydney, NSW, Australia, 3University of New South Wales, Sydney, NSW, Australia
Purpose/Objective(s): Lung cancer patients are prime candidates for adaptive radiation therapy because average GTV reductions during a treatment course exceed 40%. However, evaluating the dosimetric impact of anatomical changes to determine the need for a replan is currently labor-intensive. This process requires volumetric imaging suitable for dose calculation, segmentation of target volumes and organs at risk (OARs), and plan recalculation. Consequently, many patients who could benefit from offline adaptive radiation therapy do not receive it due to resource constraints. This work evaluates an automated software solution using daily cone beam CT (CBCT) images to reconstruct delivered dose distributions, specifically benchmarking the registration and dose calculation algorithms against AAPM TG-132 and TG-218 recommendations.
Materials/Methods: A cohort of ten patients with lung cancer was evaluated using mid-treatment CBCTs, including five treated on Elekta linacs (ethics-approved dataset; 2024-09-1157-PRE-1) and five on Varian linacs (TCIA 4D-LUNG dataset). The entire workflow was completed automatically without manual intervention. The synthetic CT (sCT) was generated by deformably registering the planning CT (pCT) to the daily CBCT via initial rigid alignment and slab-wise histogram matching, followed by a multi-resolution Demons algorithm (Thirion, 1998). Delivered dose was calculated on the sCT using a Collapsed Cone Convolution (CCC) algorithm (Ahnesjö, 1989). Following AAPM TG-132, the registration performance was quantified by the target registration error (TRE) of 16–20 anatomical landmarks per patient, labeled by a medical physicist and reviewed by a radiation therapist. The dose calculation accuracy was evaluated against reference AcurosXB distributions (Varian Eclipse v18.0, dose-to-medium) using global gamma with a 3%/2mm criteria and 10% dose threshold (AAPM TG-218). Dosimetric comparisons included D95% for target volumes, D2% for serial OARs, and Dmean for non-serial OARs.
Results: The median TRE across all patients was 2.7 mm (95% confidence limit: 4.3 mm), demonstrating registration accuracy within the 3 mm voxel dimension of the CBCT images. The CCC dose distributions showed high agreement with the reference dose distributions, with a median 3%/2mm gamma pass rate of 97.5% (95% confidence limit: 96.5%). Furthermore, 100% of target volume (D95%) and OAR metrics (Dmean, D2%) were within 5% of the prescription dose.
Conclusion: This study demonstrates an automated method for dose reconstruction on daily CBCTs for offline adaptive radiation therapy. The algorithms meet AAPM TG-132 guidelines, with a median TRE below the maximum voxel dimension (3 mm), and TG-218 tolerances, with a 3%/2mm gamma pass rate exceeding 90%. The automated approach enables routine clinical dosimetric monitoring relative to the original treatment plan, mitigating resource constraints and providing a feasible pathway for evidence-based offline adaptation.