Room: Track 5
Large-scale implementation of adaptive radiotherapy requires automated methods for reconstructing dose, including efficient quality control (QC) of dose calculation accuracy (DCA). The purpose of this work is to benchmark DCA on CBCT images and develop a QC framework using statistical process control (SPC) to monitor automated CBCT density assignments and daily reconstructions.
Daily dose reconstructions were performed in RayStation (v6.1) on 30 oropharyngeal cancer (OPC) patients treated with daily CBCT using histogram thresholding and bulk density assignment. All dose comparisons used deformed planning CTs as the gold standard, and DCA was quantified as mean and near-max (99th percentile) absolute voxel-wise dose differences. Histogram-based density thresholds for first-fraction CBCTs were defined manually (n=10, manual cohort) or using an automated RayStation algorithm (n=20, automated cohort). First-fraction thresholds were copied to subsequent CBCTs for each patient. Inter-patient control limits derived from the manual cohort were applied to monitor first-fraction errors in the automated cohort. Intra-patient control charts were generated for 10 patients with acceptable first-fraction DCA to detect errors on subsequent fractions.
Manual and automated cohort averages for the first-fraction DCA were 0.4% and 1.1% mean voxel differences, and 2.4% and 3.7% near-max differences. The inter-patient control chart detected first-fraction out-of-control signals in 16/20 patients in the automated cohort. Most signals corresponded to poorly-computed thresholds due to dental implants and CBCT artifacts. Manual intervention to redefine thresholds yielded acceptable DCA matching the manual cohort. Intra-patient control charts applied to serial CBCTs detected signals in 2/100 examined fractions, indicating excellent CBCT histogram stability.
We developed a framework for automated QC of CBCT dose reconstruction based on inter- and intra-patient SPC. Manually defined thresholds for bulk density assignment are typically required for first-fraction CBCTs in our OPC cohort. However, DCA of subsequent fractions can be maintained and monitored effectively using SPC.
Not Applicable / None Entered.