Evaluation of Deformable Image Registration-Based Dose Accumulation for Parotid Glands in Head and Neck Adaptive Radiotherapy and its Impact on Prediction of Xerostomia

Abstract

<h3>Purpose/Objective(s)</h3> Deformable image registration (DIR) so far is the fundamental tool for dose accumulation despite the noticeable errors and uncertainty. Accurate dose and NTCP evaluation are crucial in adaptive radiotherapy (ART). The current study assessed the difference of mean dose (D_mean) between the accumulation by DIR and ground truth for parotid gland and the impact on NTCP of xerostomia. We aim to propose a method using DIR for fast and accurate dose and NTCP evaluation in ART. <h3>Materials/Methods</h3> A total of 123 head and neck cancer (HNC) patients who underwent photon radiotherapy and had at least 5 weekly in-treatment re-scan CTs (rCTs) were enrolled. Parotid glands were segmented on planning CT and each rCT by the same senior radiation oncologist with more than 10 years' experience in OAR segmentation. The original treatment plan (VMAT or IMRT) was re-calculated on each rCT to obtain the actual weekly mean dose (D_mean). was performed between planning CT (reference CT) and each rCT (target CT) and then used to warp the dose distribution of each rCT backward and accumulate on planning CT to obtain the DIR accumulated D_mean (D_{mean_DIR}). The ground truth D_mean (D_{mean_GT}) was defined as the average of the actual weekly D_mean. The planning D_mean, D_{mean_DIR} and D_{mean_GT} were all translated into NTCP for xerostomia (van den Bosch et al., 2021). NTCP of D_{mean_DIR} and planning D_mean were used to develop a liner multiple regression model to predict NTCP of D_{mean_GT}. All the combinations of coefficients within the 95% confidential interval were tested, the prediction error was defined as NTCP of D_{mean_GT} - prediction of NTCP. <h3>Results</h3> There is no statistically significant difference between D_mean difference of D_{mean_GT} (D_{mean_GT}- planning D_mean) and D_{mean_DIR} (D_{mean_DIR} - planning D_mean). The NTCP difference (NTCP based on D_{mean_DIR} – NTCP based on D_{mean_GT}) of xerostomia at the 6^th month after radiotherapy was 0.01 ± 0.96% (range: -3.94% to 2.63%, assumed Grade 0 baseline xerostomia) and 0.05 ± 1.17% (range: -3.71% to 2.99%, assumed Grade 1-2 baseline xerostomia). In both models, the p value of coefficients for NTCP of planning D_mean and D_{mean_DIR} were significant and <0.01. For cases showing increased NTCP during the course of RT >3 pp (NTCP of D_{mean_GT} vs. NTCP of planning D_mean), adjusting the coefficients within the 95% confidential interval could help to predict the NTCP of D_{mean_GT} within a deviation of -1.31% to 1.89% and -1.58% to 1.99% for baseline xerostomia of Grade0 and Grade 1-2, respectively. <h3>Conclusion</h3> The dose accumulation by deformable image registration could incur a prediction error up to 3.94 pp for NTCP of xerostomia. For patients showing increased NTCP of xerostomia more than 3 pp due to dose change, our model using the NTCP based on planning and DIR accumulated mean dose of parotid glands could be used for accurate prediction of the actual NTCP with a prediction error less than 2 pp.