Prospective assessment of inter- or intra-fractional variation according to body weight or volume change in patients with head and neck cancer undergoing radiotherapy

This study aimed to prospectively investigate the association between body weight (ΔBW) or body volume variations (ΔBV) and inter- or intra-fractional variations (Δ(inter) or Δ(intra)) in patients with head and neck cancer (HNC) undergoing radiotherapy (RT). This study enrolled patients with HNC from December 2015 to December 2017. All patients underwent curative intensity-modulated RT (IMRT) either as definitive or adjuvant treatment. Six-dimensional inter- and intra-fractional variations (Δ(inter) and Δ(intra)) were obtained with ExacTrac (BrainLAB, Feldkirchen, Germany) system. BV was measured 7.5 cm cranio-caudally from the centre using cone beam computed tomography. The BW, BV, and Δ(inter) were calculated based on the value obtained on the first treatment day after each simulation. Both Δ(inter) and Δ(intra) were considered in calculating the optimal margins for planning target volume (PTV), which was calculated using van Herk’s formula. In total, 678 fractions with 39 simulations in 22 patients were analysed. The average ΔBW and ΔBV was -0.43±1.90 kg (range, -7.3 to 5.0) and -24.34±69.0 cc (range, -247.15 to 214.88), respectively. In correlation analysis, Δ(intra) was more associated with ΔBW or ΔBV than Δ(inter). Receiver operating characteristic analysis showed Δ(intra) could differentiate ΔBW from ΔBV, while Δ(inter) could not. The optimal margins for PTV considering both Δ(inter) and Δ(intra) were 3.70 mm, 4.52 mm, and 5.12 mm for the right-left, superior-inferior, and anterior-posterior directions, respectively. In conclusion, the PTV margin of 6 mm for anterior-posterior direction and 5 mm for the other directions were needed. ΔBW or ΔBV correlated with Δ(intra) rather than Δ(inter). Therefore, ΔBW or ΔBV should be assessed for accurate IMRT in patients with HNC.

±1.90 kg (range, -7.3 to 5.0) and -24.34±69.0 cc (range, -247. 15 to 214.88), respectively. In correlation analysis, Δ(intra) was more associated with ΔBW or ΔBV than Δ(inter). Receiver operating characteristic analysis showed Δ(intra) could differentiate ΔBW from ΔBV, while Δ (inter) could not. The optimal margins for PTV considering both Δ(inter) and Δ(intra) were 3.70 mm, 4.52 mm, and 5.12 mm for the right-left, superior-inferior, and anterior-posterior directions, respectively. In conclusion, the PTV margin of 6 mm for anterior-posterior direction and 5 mm for the other directions were needed. ΔBW or ΔBV correlated with Δ(intra) rather than Δ(inter). Therefore, ΔBW or ΔBV should be assessed for accurate IMRT in patients with HNC. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 Introduction A multimodal approach that includes radiotherapy (RT) is important for the successful treatment of head and neck cancer (HNC) [1]. Intensity-modulated radiotherapy (IMRT) has recently become the preferred modality for radiotherapy in patients with HNC because it offers more conformal dose distribution. IMRT could spare the parotid glands, thus reducing the incidence of xerostomia [2]. However, the steep dose gradient of an IMRT plan poses a risk of marginal miss due to set-up error [3]. To compensate for this risk, a margin from the clinical target volume (CTV) is added to the planning target volume (PTV). However, although a larger margin lowers marginal misses, it also delivers a higher dose to organs at risk. Therefore, it is necessary to determine the optimal margin required clinically by accurately measuring the error.
Errors that can cause marginal misses can be divided into two types, namely, inter-fractional or intra-fractional variations (Δ(inter) or Δ(intra)). The Δ(inter) is caused by variations of the patient's body [4] and set-up errors between each daily treatment. To correct for Δ (inter), image-guided radiotherapy (IGRT) is performed, which corrects the patient's posture pre-treatment. Meanwhile, the Δ(intra) is mainly caused by the fine movement of the patients during RT [5,6].
Although RT is beneficial for HNC, it also results in dysphagia, odynophagia, dry mouth, and a loss of sense of taste, thus often leading to weight loss [7]. Chamchod et al. [8] reported that the mean posttreatment body mass index of patients with cancer is lower than that in pretreatment (28.5 ± 4.9 kg/m 2 vs 26.2 ± 4.4 kg/m 2 in men; 27.8 ± 8 kg/m 2 vs. 26 ± 7.5 kg/m 2 in women). Furthermore, in cases of cervical lymph node metastasis, the surface of the head and neck can change during RT [9]. Shrinkage of the parotid glands during RT has also been reported [10]. As the surfaces of the head and neck change, a gap may occur between the thermoplastic mask and the skin, which may increase potential error. Lai et al. [11] reported correlations between circumferences at the level of the mastoid tip and inter-fractional variations. However, the measurement in their study was performed only twice in the RT course and was also studied by measuring the length only in two dimensions rather than measuring by body volume (BV) in three-dimensions.
In this study, we measured the variation of body weight (BW, ΔBW), BV (ΔBV), Δ(inter) and Δ(intra) with the aim to determine the relationship between body indices (ΔBW or ΔBV) and set-up errors (Δ(inter) or Δ(intra)) to ultimately suggest an optimal margin for PTV from CTV in RT for HNC.

Immobilization and simulation
DUON TM (Orfit Industries, Wijnegem, Belgium) masks were used as immobilization devices (Fig 1). Bite block was used as needed. All patients were scanned using a Brilliance Big Bore CT simulator (Philips Inc., Cleveland, OH) with a thickness of 2.5 mm. If there was an Δ (intra) error of more than 3 mm in post-treatment ExacTrac (BrainLAB, Feldkirchen, Germany), repeated simulation was performed.

Treatment planning
An experienced radiation oncologist delineated the gross tumour volume and the CTV for each patient. The PTV was generated with a margin of 5 mm in the CTV. Nine-field static IMRT was generated with inverse planning using anisotropic analytical algorithm. The Eclipse 8.6 (Varian Medical System, Palo Alto, CA, USA) was used as the treatment planning system.

Image registration and setup protocol
For this study, the kV ExacTrac system and cone beam computed tomography (CBCT) using a kV on-board-imager were used for image guidance. The patients' BW was measured during each treatment fraction. After the treatment position was set-up, the initial ExacTrac was performed. Image registration was performed using a digitally reconstructed radiograph (DRR) from the CT simulation image as described in our previous study [12]. Image registration was performed automatically based on the bony anatomy of the mandible and cervical spine using the region of interest (ROI) function with spyglass mode, checked by an experienced therapist and then confirmed by an experienced radiation oncologist with an offline review. Then, the correction based on the results of initial ExacTrac was applied to the 6D robotic couch system (BrainLAB, Feldkirchen, Germany). Repositioning was performed as necessary. After correction, pre-treatment ExacTrac and CBCT were performed. Correction via CBCT or pre-treatment ExacTrac was not allowed. Post-treatment ExacTrac was then performed. If an error exceeded 3 mm in any direction of the post-treatment ExacTrac, we performed repeated simulation.

Treatment delivery
All patients received IMRT treatments using a Novalis Tx (Varian Medical System, CA, USA) linear accelerator machine with HD-120 multi-leaf collimator. Also, photon energy of 6 MV for the radiation dose was delivered to tumour with a dose rate of 600 monitor units per minute.

Statistics
We used Pearson's correlation analysis to determine the correlation between ΔBW and ΔBV. We also used Pearson's correlation analysis for ΔBW or ΔBV and Δ(inter), Δ(intra), absolute value of Δ(inter), and absolute value of Δ(intra). Furthermore, receiver operating characteristic (ROC) analysis was performed to determine the optimal points of ΔBW or ΔBV for treatment in less than 3 mm of Δ(inter) or 2 mm of Δ(intra). Box plot of Δ(inter) or Δ(intra) for each group was created based on ΔBW of 1 kg. All analyses were performed using SPSS version 25.0 (IBM Inc., Armonk, NY, USA).

Recipes for optimal PTV margins
Margin recipes proposed by Stroom et al. [13] and van Herk et al. [14] were used to calculate the appropriate margin for PTV based on systematic SD (∑) and random SD (σ). Margins were calculated considering both Δ(inter) and Δ(intra). The total ∑ and σ were calculated according to the following equations:

Patient characteristics
Twenty-two patients were enrolled. Consequently, 678 fractions with 39 simulations were analysed. The patient characteristics are summarized in Table 1. Two patients (patients numbered 18 and 22) discontinued treatment due to sepsis and patient refusal. During the course of RT, In this study, 6 of the 22 enrolled patients attached with bite block. As the use of bite block may introduce potential bias, we performed additional analyses comparing these patients to those who did not attach with bite block. Because the Kolmogorov-Smirnov test revealed that the data for ΔBV, ΔBW, and Δ (intra) and Δ (inter) were not normally distributed among patients who used bite block (p <0.05), the Mann-Whitney U nonparametric test was used for An assessment of setup error according to body change in head and neck cancer patient these variables. When bite block was used, ΔBW was -0.15 ± 1.99 kg; when bite block was not used, ΔBW was -0.6 ± 1.83 kg. Weight loss was greater in patients who did not use bite block (p = 0.001). However, there was no significant difference in ΔBV between patients who used or did not use bite block (-12.2 ± 57.46 cc versus -29.26 ± 72.67 cc, p = 0.072).

ROC analysis
ROC analysis based on Δ(inter) �3 mm and Δ(intra) �2 mm of RL, AP, SI, 3D vector, and any directions (when any one of the RL, SI, or AP was outside the specified tolerance) was performed (Table 7) (Fig 5A-5C).

Comparison of 1 kg group of ΔBW
A box plot of the ΔAP(intra), Δ3D(intra), and Δpitch(intra) grouping in groups of 1 kg of ΔBW are shown in Fig 6A-6C. These graphs showed larger variations in patients with weight loss of more than 3 kg. No profound differences were noted in other directions of Δ(intra) and all directions of Δ(inter). Table 8 summarizes the optimal margins for the PTV based on systematic SD (∑) and random SD (σ) as previously proposed [13,14]. Based on Stroom's formula, the calculated margins considering Δ(inter) ranged from 2.87 mm to 3.99 mm, and the margins considering both Δ (inter) and Δ(intra) ranged from 3.04 mm to 4.19 mm. Based on van Herk's formula, the calculated margins are 3.5-4.88 mm and 3.7-5.12 mm, respectively.

Repeated simulation due to the post-treatment ExacTrac exceeding 3 mm
Repeated simulation was performed when the post-treatment ExacTrac exceeded 3 mm in RL, SI, or AP. Patient number of 8 and 11 required repeated simulations after the 15 th and 17 th fractions, respectively (Fig 7A and 7B). For both of these patients, there was no planned repeat simulation.

Optimal margins for the PTV from the CTV
In this study, the ExacTrac system and CBCT were used to calculate the optimal margins from the CTV to the PTV. However, Qi et al. [15] have reported that daily random setup errors and the CTV-to-PTV margins for treatment of head-and-neck cancer were highly affected by  An assessment of setup error according to body change in head and neck cancer patient IGRT modalities and image quality. To identify the differences between image systems, in our previous study [16], we reported that the clinical setup discrepancy between the ExacTrac system and the CBCT for the intracranial stereotactic radiosurgery was <1.01 mm and <0.82f or online matching. Assessment of intra-fractional motion with pre-and post-kV imaging as a surrogate of intra-fractional motion is an important issue. However, many studies have previously used multi-modalities such as computed tomography (CT), CBCT, magnetic resonance imaging (MRI), and kV images as surrogates of intra-or intra-fractional motion [4][5][6][7][8][9][10][11][12]. These studies have assumed that the evaluation of motion trajectory at specific time points, such as before and after treatment, represents inter-or intra setup variation in radiotherapy. Chan et al. [17] has pointed out that continuous fluoroscopic imaging can trace the tumour motion trajectory throughout the radiotherapy treatment, but the patient may receive additional undesired radiation dose. Therefore, we assumed that the kV images before and after treatment sufficiently represented intra-fractional variations.

ΔRL(mm) ΔSI(mm) ΔAP(mm) Δpitch(˚) Δroll(˚) Δyaw(˚) Δ3D(mm)
The criteria for optimal margins in PTV have not been established. In general, the Radiation Therapy Oncology Group protocols recommend a minimum of 5 mm margin around the CTV in all dimensions for HNC patients who did not undergo IGRT [18]. The calculated margins of our study considering only Δ(inter) for the PTV based on Stroom's and van Herk's formula [13,14] are generally consistent with those of previous studies. In previous studies, the optimal margins calculated based on van Herk's formula were 3.4-3.8 mm, 3.8-4.8 mm, and 3.7-4.4 mm in the RL, SI, and AP directions, respectively [19][20][21] (Table 9). Considering Δ (inter) only, a 5 mm margin for PTV seems to be sufficient in all dimensions. The calculated margins based on van Herk's formula considering both Δ(inter) and Δ (intra) were analogous to the results obtained by Cacicedo et al. [22] in the RL (3.7-4.2 mm) and SI (3.9-4.5 mm) directions, but there was a difference of approximately 1 mm in the AP direction (4.1 mm in the study of Cacicedo et al. [22] vs 5.12 mm in our study, Table 10). In our study, ExacTrac was used, while a previous study [22] measured the error using only the AP and lateral field via an electronic portal imaging device. ExacTrac provides 6D information on the error, which is calibrated using the 6D couch, allowing for more accurate measurements of Δ(inter) and Δ(intra).
In addition, a box plot of ΔAP(intra) grouping in groups of 1 kg suggests that weight loss >3 kg may result in larger Δ(intra). This result suggests that repeated simulation may be needed when the weight loss is over 3 kg. Because Δ(intra) is an important issue in the RT for HNC patients, further studies should be considered. Recently, Navran et al. [27] reported that reducing the CTV-PTV margin from 5 mm to 3 mm, combined with daily CBCT-guided VMAT, reduced the radiation-related toxicity. Thus, in order to reduce the margin, it is important to maintain body weight during radiotherapy.
The absorbed dose of the image guidance system by multi modalities should be confirmed in image registration and setup protocol. Because single verification with X-ray tube 1 and 2 with ExacTrac system is 1 mSv [12], the overall absorbed dose for the ExacTrac system is 4 mSv with 2mSv for inter-fractional variation, and 2mSv for intra-fractional variations. This dose is lower than the dose of CBCT (14 mSv).

Study limitations
First, BV measurement using CBCT has some uncertainties. Compared to conventional CT, distinct artefacts present in CBCT could impact BV measurement, although we believe this risk disappears due to the large number of fractions. Second, because BW measurement and CBCT were not performed on all patients, only 71.5% of ΔBW and 95.4% of ΔBV were An assessment of setup error according to body change in head and neck cancer patient   analysable. In addition, pre-treatment ExacTrac was not performed in one fraction because of a mechanical problem. Moreover, post-treatment ExacTrac was missed in 8 of the 678 fractions. Third, weight loss was smaller than expected. The average ΔBW was -0.43±1.90 kg in all fractions, and ΔBW was increased in 37.7% of the analysable fractions. Also, we performed repeated simulation when the post-treatment ExacTrac showed over 3 mm of variation which reflect Δ(intra). Thus, excessive Δ(intra) was suppressed. As such, it was difficult to observe any significant set-up error.

Conclusions
The PTV margins of 6 mm for AP direction and 5 mm for the other directions were needed to include 95% of both inter-and intra-fractional set-up variations. Further, BW and BV were correlated with set-up variations. Specifically, intra-fractional variations were positively correlated with changes in BW or BV. Therefore, assessments of BW and BV should be considered for accurate IMRT in patients with HNC.

Acknowledgments
This work was supported by the 2018 Yeungnam University Research Grant.

Author Contributions
Conceptualization: Jae Won Park.
Data curation: Seong Hoon Kim, Se An Oh, Jae Won Park.