Parotid gland radiation dose‐xerostomia relationships based on actual delivered dose for nasopharyngeal carcinoma

Abstract Xerostomia induced by radiotherapy is a common toxicity for head and neck carcinoma patients. In this study, the deformable image registration of planning computed tomography (CT) and weekly cone‐beam CT (CBCT) was used to override the Hounsfield unit value of CBCT, and the modified CBCT was introduced to estimate the radiation dose delivered during the course of treatment. Herein, the beams from each patient's treatment plan were applied to the modified CBCT to construct the weekly delivered dose. Then, weekly doses were summed together to obtain the accumulated dose. A total of 42 parotid glands (PGs) of 21 nasopharyngeal carcinoma patients were analyzed. Doses delivered to the parotid glands significantly increased compared with the planning doses. V20, V30, V40, Dmean, and D50 increased by 11.3%, 28.6%, 44.4%, 9.5%, and 8.4% respectively. Of the 21 patients included in the study, eight developed xerostomia and the remaining 13 did not. Both planning and delivered PG Dmean for all patients exceeded tolerance (26 Gy). Among the 21 patients, the planning dose and delivered dose of Dmean were 30.6 Gy and 33.6 Gy, respectively, for patients with xerostomia, and 26.3 Gy and 28.0 Gy, respectively, for patients without xerostomia. The D50 of the planning and delivered dose for patients was below tolerance (30 Gy). The results demonstrated that the p‐value of V20, V30, D50, and Dmean difference of the delivery dose between patients with xerostomia and patients without xerostomia was less than 0.05. However, for the planning dose, the significant dosimetric difference between the two groups only existed in D50 and Dmean. Xerostomia is closely related to V20, V30, D50, and Dmean.


| INTRODUCTION
Oral complications induced by radiotherapy are universal in the treatment of head and neck carcinomas (HNC). Xerostomia results from the deleterious effect of radiation, and is the most common persistent oral sequela for HNC patients who receive therapeutic doses. 1 It occurs with difficulties in swallowing and speaking, loss of taste, and dental caries, which have a detrimental impact on the quality of life. 2 The occurrence of xerostomia is related to several factors such as the radiation dose, the volume of irradiated tissue, and the use of concurrent radiation sensitizing and chemotherapy drugs. Much research has been performed on delivering radiation to normal tissue to spare patients from the development of complications. Intensity-modulated radiation therapy (IMRT) has been considered as the standard therapeutic technique of HNC because of its highly conformal, modulated techniques. 3,4 It permits the delivery of a high dose to the target volume and spares the surrounding critical structures. Several studies have noted that IMRT has advantages over three-dimensional conformal radiotherapy in terms of sparing the parotid gland (PG), and it can improve the quality of life for HNC patients. 5,6 However, position variations and anatomical changes over the course of treatment limit the benefits of IMRT and lead to a clinically significant dose difference between the initial plan and delivery.
Previous studies have shown that anatomy changes may cause more dose deviations in organs at risk (OARs) than in target volumes. [7][8][9][10] It is true that changes in anatomy were found among most of the nasopharyngeal carcinoma (NPC) patients. 11 Generally, bodyweight loss, primary tumor shrinkage, and PG volume reduction could induce a PG dose change and increase the risk of xerostomia.
In order to correct these deviations, adaptive radiotherapy (ART), online re-optimization, or offline replanning is recommended. 12,13 Castelli et al. investigated the impact of ART to spare the PG and decrease the risk of xerostomia using weekly computed tomography (CT), and included the PG overdose, the benefit of ART, and anatomical markers related to dose differences between the planning and accumulated dose (with/without replanning). 2 Although the time and the optimal number of replanning incidents remain unclear and the strategies of replanning remain uncertain, the benefits of ART have previously been demonstrated. 14 However, it is not likely that patients can obtain CBCT scans for every fraction during the entire treatment, and not every patient will benefit from ART. Thus, identifying patients earlier by selection criteria to spare OARs and further decrease the risk of complication will be beneficial to both patients and radiation departments. Previous studies selected patients by determining some criteria that would allow them to benefit from ART or replanning. 4,15,16 In clinical practice, CT is the only imaging modality that can be used for treatment planning. Cone-beam CT (CBCT) integrated into linear accelerators cannot be used as a planning imaging modality because of its inferior image quality, poor electron density accuracy, scatter, and motion artifacts. However, CBCT scans possess the position and shape information of the target and organs at risk in real time. Daily or weekly on-board CBCT is used to assist patient setup. In addition, CBCT data can also potentially be used for dose reconstruction with the electron density calibrated. Yang  In our previous study, we investigated dosimetric variations of the liver using deformable registration of planning CT (pCT) and CBCT and found that there was an increase in D 50 and D mean when compared with the planning dose. 19 Additionally, a significant dosimetric difference between patients with and without radiationinduced liver disease (RILD) has been reported. Charlotte et al.
stated in their review that only a few articles reported the clinical relevance of dosimetric changes in terms of complications of head and neck cancer patients. 16 As for dosimetric variation, the delivered dose may not correspond to the planned dose. Several studies investigated PG dose variation and found great dosimetric changes. 10,12,20 However, inaccuracies in the calculation of dosimetric change should not be ignored.
One of the reasons leading to the inaccuracies is that in most studies, treatment plans used to calculate dose distributions were constructed on CT acquired a few days before radiotherapy, such as repeated planning CT. Patients experience anatomical changes during the treatment course, and consequently, a plan based on pretreatment CT images cannot precisely reflect the actual dose distribution during fractionated IMRT. In that case, the correlation between clinical outcomes and dosimetric changes might not be accurate. Hence, a real-time imaging modality such as CBCT is a better choice for estimating the delivered dose and investigating the relevance of dosimetric changes and complications.
Xerostomia might not occur in salivary gland tumors because the administered radiation is usually restricted to the ipsilateral gland.
However, for the nasopharynx, the parotid glands are usually affected by radiation, and severe and permanent xerostomia may result. In this study, 21 NPC patients were included to investigate the association between dosimetric factors and xerostomia. We hypothesize that the actual delivered dose based on CBCT correlates more closely with the development of xerostomia than the planning dose, and some dosimetric characteristics between patients with xerostomia and patients without xerostomia should be different. This study aims to validate the hypothesis by using the mCBCT approach.  Table 1. The study was approved by the Shandong Cancer Hospital ethics committee. All patients provided written informed consent.

2.B | Treatment and planning
All patients underwent simulation on Philips CT Big Bore, immobilized in a supine position with a thermoplastic mask covering the head, neck, and shoulders. Intravenous contrast-enhanced CT using 3 mm slice thickness was carried out, ranging from the vertex to the manubria sternal joint for planning. These data were transferred to the treatment planning system (Pinnacle, version 9.2 to 9.8). Targets were delineated by clinicians with the assistance of a combination of CT and MRI. Magnetic resonance scans were obtained using the same position of simulation. The delineation of organs at risk was completed by physicians. All patients' radiation doses were planned and delivered using inverse IMRT. The prescribed radiation dose was 70 Gy at 2 Gy per fraction delivered over 6-7 weeks (5 or 6 fractions a week). The objective parameter used in IMRT optimization for the PG was at least one PG D mean < 26 Gy or D 50 < 30 Gy.
All patients underwent CBCT scans with linear accelerators (Varian Trilogy) once a week to correct setup errors before radiotherapy.
CBCT scanning was performed with 2.5 mm slice thickness. These CBCT scans were noted as CBCT week 1, week 2, week 3, and so on. Patients included in the study received chemotherapy with cisplatin and/or tegafur.

2.C | Dose construction
A plan based on mCBCT was generated to calculate the delivered dose.
First, weekly CBCT and pCT were registered using rigid registration, followed by deformable image registration (DIR) to obtain the correct Hounsfield unit (HU) value of CBCT. Then, beam configurations and dose constraints of the initial plan were reapplied to mCBCTs. The target volume and OAR delineations upon pCT were automatically propagated onto mCBCT using the deformation vector field resulting from the registration as described above. All automatically propagated delineations were checked by the same proficient physicist. Delineations would be revised if the mapping structures were not in accordance with the anatomy on the mCBCT unless the results satisfied the physicist. Last, the dose distribution was recalculated.

2.D | Gradient-based deformable image registration
3D/2D registration methods can be classified into three categories: extrinsic, intrinsic, or calibration-based. 21 The intrinsic methods can be further divided into two main categories, feature-and intensitybased, and it is accepted that the latter has an important advantage over the former in that all available information can be utilized in the images. However, it is possible that registration would fail because of the inaccuracy of the CBCT intensity. Scatter is one of the main reasons for intensity inaccuracy. Many prior studies showed that the scattering of CBCT results in a low-frequency signal. That is to say, in the scatter artifact images, the shapes of almost all inner and outer object boundaries, such as boundaries between bone, fat, muscle, and air, can be perceived. 22 Based on the characteristics of scatter, we proposed the gradient-based free-form deformation algorithm (GFFD). 23 The GFFD algorithm measures the similarity by using 3D gradient vector fields. The local polynomial approximation-intersection of confidence intervals algorithm is performed to accommodate the image sampling anisotropy. A "bi-directional" force along with an adaptive force strength adjustment is introduced to accelerate the effect of the convergence process. These strategies are expected to decrease the effects of the inconsistent intensities.

2.E | Dose calculation and accumulation
The planning dose was calculated at the time of treatment planning.
The accumulated delivered dose was derived from the weekly CBCTs. Firstly, the weekly mCBCT was aligned and registered with  and D mean of the PG. Figure 1 shows the derivation of the total planning dose and accumulated delivery dose. Figure 2 shows the forward and inverse registration steps.

2.F | Statistical analysis
The paired samples t-test was used to obtain the association between the planning dose and the actual delivered dose. The independent samples t-test was used to determine whether there is a statistically significant dose difference between patients with xerostomia and patients without xerostomia. All analyses were performed using IBM SPSS version 23.0 software. A P-value less than 0.05 was considered significant.  with CBCT, and it is comparable to that of pCT. There is less noise on the mCBCT as compared with CBCT. More importantly, the electron density of mCBCT is calibrated. Herein, accurate dose construction is allowed.

3.A | Imaging and automatic propagation
The inverse deformation field of registration mapped the contours on CBCT from pCT. Figure 5 illustrates the automatic propagation of the parotid gland of an NPC patient. The last row of Fig. 5 shows the deformed contours overlaid on the original CBCT images.
As can be seen, the automatically generated contours on mCBCT match the structures on CBCT well, which allowed us to perform the delivered dose calculation.   Fig. 6.

| DISCUSSION
The algorithm of DIR has an impact on the results of this study because the HU value of mCBCT overridden by DIR would influence the accuracy of dose construction. Additionally, the DIR algorithm determines the accuracy of propagation of structural contours. We have researched the deformable registration algorithm over the last few years. 24,25 We also researched automatic contouring based on segmentation for radiotherapy treatment planning. [26][27][28] We found that intensity-based deformable registration algorithms are susceptible to distorting tissues because the presence of inconsistent intensities between CT and CBCT could cause significant registration errors. The DIR algorithm performed in this study was presented in our previous work, and was tested using phantom and clinical data, which demonstrated the significance of the registration method. 23 We assessed the volume overlapping using the Dice Similarly coefficient (DSC) in that study.
The DSC value between the contours propagated by the GFFD algorithm and the contours edited by the oncologist was 96% for the parotid gland.
Many studies have been devoted to investigating the side effects of radiotherapy in head and neck cancer. This is mainly because OARs and tumor target volumes of HNC are close to each other. 16 The parotid gland is the most studied organ at risk in head and neck Generally, PG volume will decrease after radiotherapy. Previous studies suggested that, on average, the PG volume reduction rate was 26 AE 11%. 16 In our study, the PG volume reduction rate was 13.5% and 13.2%, respectively, for patients with xerostomia and patients without xerostomia (Table 5). Although there is little difference in the PG volume reduction rate between patients with xerostomia and patients without xerostomia, the dose deviations between the two groups greatly differ. More study is needed to explore this issue.
It is worth mentioning that although the D mean value of several patients without xerostomia was over the tolerance of 30 Gy, some were even larger than that of patients with xerostomia but they did not develop xerostomia (Fig. 7). OARs. Yet, it seems that not all patients are able to benefit from ART because it is challenged by time-consuming procedures and requires extra resources. Additionally, the selection criteria for identifying patients who may benefit from ART are ambiguous, which obstructs the implementation of ART. The replanning strategies, timing, and optimal number of treatments remain unclear so far. Many researchers are investigating these issues, and some have made new progress. 15,37 We firmly believe that it will lead us to new important discoveries if these studies can be performed using CBCT.

| CONCLUSION
Doses actually delivered to the parotid glands significantly increased compared with planning doses. Statistically significant differences in V 20 , V 30 , D mean , and D 50 for parotid glands were found between patients with xerostomia and patients without xerostomia.