Session Item

In adaptive proton therapy workflows, robustness is often monitored using weekly repeat CTs (rCTs). Alongside the rCTs, daily in-room Cone-Beam CT (CBCT) images are acquired for pre-treatment position verification. While CBCT images are not suitable for direct proton dose calculations, as they suffer from severe imaging artifacts, they can be converted into higher quality synthetic CT (sCT) images using neural networks. The image quality of CBCT-based sCTs has previously been shown as similar to rCT quality, but the dosimetric accuracy of sCTs in a clinical setting is yet to be determined.

The aim of this work was to compare the daily sCT images, generated by a neural network, to the planning CT (pCT) and rCTs of head and neck (HN) cancer patients to investigate the dosimetric accuracy of CBCT-based sCTs, towards treatment robustness evaluations and online adaptive proton therapy workflows.

A dataset of 57 HN cancer patients previously treated with proton therapy at our center was used to generate synthetic CT images from daily CBCTs, using a previously developed and trained UNet-like deep convolutional neural network. The clinical treatment plans of the patients were used to recalculate the proton dose on the weekly rCTs and daily sCTs. The dose calculations were then compared to dose calculations on the pCT with the relative dose difference between the calculations in specific regions of interest (ROIs) as the metric. The investigated ROIs were the clinical target volumes (CTV) and eight organs-at-risk (OARs) used to calculate the NTCP models scores, like the submandibular and parotid glands (see figure 1). A second comparison was done by calculating the Normal Tissue Complication Probabilities (NTCP) for dysphagia and xerostomia, on the rCTs and the same-day sCTs to investigate the clinical relevancy of possible dose differences.

The relative mean dose difference found in the CTV 70 Gy was 0.1±0.3% in the pCT/rCT and -0.2±0.9% in the pCT/sCT group. Figure 1 shows the relative difference per OAR, where the largest disagreement in mean relative difference is found in the left parotid gland with 1.6±9.6% (pCT/rCT) and 6.9±11.6% (pCT/sCT). The best agreement is found in the superior pharyngeal constrictor muscle with mean relative differences of 1.8±10.9% (pCT/rCT) and 4.0±12.2% (pCT/sCT). Figure 2 shows the NTCP score difference between the rCT and same-day sCT calculations. The mean difference of the overall NTCP score was found to be -1.6±2.8%. The shifts towards positive dose difference in the OARs, especially the submandibular and parotid glands, complies with the minor NTCP score increase.

This study showed the potential of sCTs for dose calculations in daily treatment robustness evaluation and online adaptive proton therapy of HN cancer patients. In comparison to rCTs, only minor differences were observed. Further investigation is required to identify where the differences originate from and if they are relevant for clinical decision making.