ART for head and neck patients: On the difference between VMAT and IMPT.

Abstract Anatomical changes in the head-and-neck (H&N) region during the course of treatment can cause deteriorated dose distributions. Different replanning strategies were investigated for volumetric modulated arc therapy (VMAT) and intensity-modulated proton therapy (IMPT). Material and methods. For six H&N patients two repeated computed tomography (CT) and magnetic resonance (MR) (CT1/MR1 at week 2 and CT2/MR2 at week 4) scans were acquired additionally to the initial planning CT/MR. Organs-at-risk (OARs) and three targets (CTV70Gy, CTV63Gy, CTV56Gy) were delineated on MRs and transferred to respective CT data set. Simultaneously integrated boost plans were created using VMAT (two arcs) and IMPT (four beams). To assess the need of replanning the initial VMAT and IMPT plans were recalculated on repeated CTs. Furthermore, VMAT and IMPT plans were replanned on the repeated CTs. A Demon algorithm was used for deformable registration of the repeated CTs with the initial CT and utilized for dose accumulation. Total dose estimations were performed to compare ART versus standard treatment strategies. Results. Dosimetric evaluation of recalculated plans on CT1 and CT2 showed increasing OAR doses for both, VMAT and IMPT. The target coverage of recalculated VMAT plans was considered acceptable in three cases, while for all IMPT plans it dropped. Adaptation of the treatment reduced D2% for brainstem by 6.7 Gy for VMAT and by 8 Gy for IMPT, for particular patients. These D2% reductions were reaching 9 Gy and 14 Gy for the spinal cord. ART improved target dose homogeneity, especially for protons, i.e. D2% decreased by up to 8 Gy while D98% increased by 1.2 Gy. Conclusion. ART showed benefits for both modalities. However, as IMPT is more conformal, the magnitude of dosimetric changes was more pronounced compared to VMAT. Large anatomic variations had a severe impact on treatment plan quality for both VMAT and IMPT. ART is justified in those cases irrespective of treatment modalities.

Intensity-modulated radiotherapy (IMRT) is the current standard of care for locally advanced headand-neck (H & N) cancer. The advantage of IMRT is its ability to tailor the dose around the target volume in a highly conformal manner and therefore improve organs-at-risk (OARs) sparing [1]. Proton therapy can even further improve treatment plan quality and thus outcome. The inverted depth dose characteristics of protons allow reducing especially the low to medium dose volumes and as a result OAR doses [2,3]. Several theoretical studies explored the advantages of protons over photons for H & N tumors [4 -6]. However, to prove the superiority of proton beam therapy certainly more prospective clinical studies are required.
One of the great challenges for fractionated radiotherapy are anatomical changes that can occur during treatment. For H & N cancer patients signifi cant shrinkage of the target volume has been reported [7]. Other potential anatomic changes are weight loss, reduction in swelling or parotid gland shrinkage [8 -10].
The clinical implementation of image-guided radiotherapy (IGRT) enabled to monitor and quantify these anatomical changes. Moreover, IGRT combined with IMRT contributed to improve local tumor control and to reduce acute and late complications, e.g. xerostomia or impairment of laryngeal function in case of H & N treatments [11,12]. Adaptive radiotherapy (ART) is expected to bring additional benefi ts. Repeated imaging together with deformable registration (DIR) and dose accumulation will fi nally help to refi ne current knowledge on dose response of OARs. Several studies have been published on treatment adaptation and replanning strategies either for IMRT or proton beam therapy [7,13 -16]. The infl uence of anatomical changes for proton therapy is more pronounced due to range uncertainties [13,17,18]. According to the authors knowledge there is no study that compares changes of photon and proton dose distributions as a consequence of anatomic changes. Performing ART studies is not straightforward for several reasons, e.g. repeating computed tomographys (CTs) during the course of therapy is not commonly performed, the workload increases with ART in clinical routine and there are limitations and uncertainties of DIR algorithms. Although in photon beam therapy cone beam CTs (CBCT) have become an integral part of today ' s linear accelerators and enable routine IGRT and even dose adaptation. However, standard CBCT are not suffi cient to recalculate the dose in proton therapy. Conventional diagnostic CTs have to be used unless CBCT is further developed, e.g. via dual energy options.
The scope of this study was to compare the need for replanning and ART strategies for H & N cancer patients between advanced photon (VMAT) and proton treatments (IMPT).

Image acquisition and volume delineation
Serial imaging data sets of six patients suffering from locally advanced H & N cancer of the hypopharynx or the oropharynx were utilized for this study. All patients were treated previously with IMRT and underwent chemotherapy prior to radiation therapy. In addition to imaging for treatment planning (CT 0 and MR 0 ), repeated CT (CT 1 , CT 2 ) and magnetic resonance (MR) scans (MR 1 , MR 2 ) were performed at weeks 2 and 4, respectively, for each patient.
All MR images were used for structure segmentation and were rigidly registered (RIR) to the respective CT for dose calculation, i.e. the GTV and OARs were delineated on MR images and later transferred to CTs. This holds for both imaging at the time of treatment planning and for biweekly examinations. One experienced radiation oncologist delineated all regions of interest (ROI) to eliminate inter-observer variability. To account for microscopic spread the GTV was isotropically expanded by 10 mm to defi ne CTV high_risk . Additionally, nodal involvement and potential risk regions were defi ned and an intermediate as well as a low risk CTV were created . Thus, three separate PTVs (5 mm margin) were considered and for each a different dose level was prescribed [19].

Treatment planning strategies
For all patients simultaneously integrated boost (SIB) plans were created using VMAT and IMPT. The fi rst plan was based on the planning CT 0 , and replanning was done based on CT 1 as well as CT 2 . SIB plans consisted of three dose levels and were assumed to be delivered in 35 fractions. Seventy gray was prescribed to the high risk planning target volume (PTV 70Gy ), 63 Gy to the intermediate risk PTV (PTV 63Gy ) and 56 Gy to the low risk PTV (PTV 56Gy ). Treatment plans were considered acceptable if 95% of the prescribed dose covered at least 95% of the respective PTV. For both modalities, OAR dose limits were based on the in-house applied IMRT protocol, where D 2% for spinal cord and brainstem was 45 Gy and 54 Gy, respectively. Mean doses to larynx were required to be lower than 50 Gy and at least one parotid should receive a D mean of less than 26 Gy. However, for IMPT plans, the OARs doses were pushed as low as possible while maintaining sufficient target coverage.
For VMAT the treatment planning system (TPS) MONACO (V. 3.2; Elekta, USA) was used. Based on our clinical experience and practice VMAT plans with two full arcs (2 ϫ 360 ° ) were created, assuming 10 MV photon beams. The maximum number of control points was 150 with a minimum segment width of 0.5 cm. The dose was calculated using a Monte Carlo (MC) algorithm with a standard deviation of 3% on a grid size of 3 ϫ 3 ϫ 3 mm 3 . Proton plans were generated with the TPS XiO (V4.41; CMS/Elekta, UK), assuming an RBE of 1. 1. In order to increase the robustness of the proton plans, four fi elds were applied (gantry angles of 40 ° , 140 ° , 220 ° , 320 ° ) to avoid air cavities and to have no beams passing directly through shoulders. Dose calculation was performed with a pencil beam algorithm on a grid size of 3 ϫ 3 ϫ 3 mm 3 . The initial spot sigma, defi ned in air, was 3 mm. The lateral spot spacing of the pencil beams was 5 mm and the distance between two layers was a magnitude of a single Bragg peak (width at 80% of the peak maximum) at the respective depth. These beam parameters were based on found online at http://informahealthcare.com/doi/ abs/ 10.3109/0284186X.2015.1028590).

Evaluation of results and statistics
All results were evaluated for biweekly and dose accumulated scenarios as described above. For the biweekly dose assessment D 2%(biweekly) and D mean(biweekly) values were extracted and compared for OARs, for each patient. Then median values and respective ranges were calculated based on the six patients. The same procedure was applied to D 2%(biweekly) and D 98%(biweekly) values for all three PTVs.
A similar approach was used for the accumulated scenario, where the same dose indices were extracted for the initial plan, adapted and non-adapted VMAT and IMPT plans. For statistical analysis a non-parametric Wilcoxon Signed-Rank test (p Ͻ 0.05) was performed for each dosimetric parameter when comparing the adapted and non-adapted approach. However, due to small sample size the power of the test might be low, therefore rather on population, the focus is placed on the individual cases throughout the paper. Figure 1 summarizes the interpatient variation of GTV and CTV volumes defi ned on MR 0 and the magnitude of the target volume shrinkage based on MR 1 and MR 2. The dosimetric analysis of the initial VMAT and IMPT plans recalculated on the repeated CTs revealed target and OAR dose variations during the course of therapy. For those three patients with rather small tumor shrinkage (see patients 1, 4, 6 in Figure 1) target coverage of VMAT plans was acceptable at weeks 2 and 4, i.e. V 95% was still Ն 95%, with OAR doses still below the constraints, thus replanning was considered unnecessary. Nevertheless, the adaptation procedure was performed, in order to see if better OARs sparring can be achieved. That was not the case for IMPT plans where coverage dropped signifi cantly. For patient 2 with the largest anatomical variations V 95% was only around 30%. IMPT treatment plans had to be adapted for all patients. Figure 2 shows dose distributions for both treatment modalities of the initial, non-adapted and adapted plan for two patients. The patient on the left in Figure 2 showed pronounced target volume shrinkage, while the patient on the right showed rather small volume changes. The corresponding quantitative dosimetric data is presented as dose volume histograms in Figure 3.

Interpatient variations
The larger anatomic variations caused an increase in the spinal cord dose for the non-adapted plan with clinically adequate values and reported in previous studies [19,20].
Both VMAT and IMPT plans were planned by the same experienced medical physicist, trained in both photon and particle therapy planning, in order to avoid interobserver differences in plan quality assessment.

Assessment of the infl uence of anatomical variations on dose distributions
The study was divided into two parts: 1) assessment of the biweekly doses; and 2) assessment of the total " accumulated " doses for adaptive and non-adaptive scenarios.
Biweekly dose assessment . In this approach only fractions of the total dose were assessed on the planning and biweekly CTs, in order to be independent of DIR imperfections. The initial plan was recalculated on the CT 1 and CT 2 , using the same isocenter, to evaluate how anatomical changes infl uenced the dose distribution. The biweekly doses were assumed to represent one third of the treatment and no dose accumulation was performed at this stage. Therefore, the initial plan based on CT 0 corresponded to the dose delivered during weeks 1 and 2, then the initial plan recalculated on the CT 1 corresponded to weeks 3 and 4 and fi nally to represent the dose delivered on weeks 5 and 6, the initial plan was recalculated on the CT 2 . However, biweekly dose evaluation represents only a " snapshot " of the current situation visualized on the respective CTs. In order to fully understand the effects of the anatomical variation on the total dose distribution, more sophisticated measures had to be applied.
Accumulated dose assessment . In the accumulated dose assessment two scenarios were considered. In the non-adapted scenario, where no replanning was performed, the initial plan was applied for the whole course of the treatment but anatomic variations were taken into account by dose re-calculations on CT 1 and CT 2 . In the adapted scenario, new treatment plans were created based on CT 1 and CT 2 . In both scenarios DIR was performed for CTs in a manually defi ned ROI using DIRART, a Matlab ® based environment with an implemented Demon algorithm [21]. The manually defi ned ROI was used in order to focus DIR only on the volume of interest, which was large enough to contain all target volumes and OARs, however axial slices were not trimmed. The deformation vector fi eld was subsequently used as basis for dose accumulation for both scenarios. The detailed workfl ow of the dose accumulation procedure is illustrated in Supplementary Figure 1 (to be 53 Gy values around the spinal cord for VMAT. Moreover, for IMPT large parts of the parotid gland received much higher doses compared to the initial plan. Adaptation of the plan helped to reduce the dose to spinal cord and parotids for both modalities (Figure 2a and b).
Looking at the patient with the small volume shrinkage, no signifi cant differences were found between the initial planned dose distributions and non-adapted doses for VMAT (Figure 2c). However, for IMPT even small volume changes resulted again in large dose deviations. The target dose contained many hot spots, even above 77 Gy. Higher doses in the non-adaptive approach were also found for the contralateral parotid gland and the larynx, up to 74 Gy and 76 Gy, respectively (see Figure 3).

Dosimetric variations (biweekly dose evaluation)
Evaluation of biweekly doses showed an increased exposure of OARs for both VMAT and IMPT compared to the initial plan. Table I  For all OARs an increase of D 2%(biweekly) was observed. Median D 2%(biweekly) of the brainstem increased from 11.2 Gy to 14.9 Gy for VMAT, while for IMPT only an increase of 1.4 Gy was found. Similarly, D 2%(biweekly) of the spinal cord increased by 1 Gy for VMAT and even by 3.3 Gy for IMPT. D mean(biweekly) increased as well for parotid glands and larynx. For both modalities median D mean(biweekly) increased by 2 Gy and 3 Gy in weeks 3 -4 and 5 -6, respectively.
For the target volumes, with VMAT no difference was observed in D 2%(biweekly) . IMPT was much more sensitive and D 2%(biweekly) increased up to 3 Gy in following weeks. Differences in D 98%(biweekly) were again more pronounced for IMPT compared to VMAT with IMPT median D 98%(biweekly) decreased by 2.5 Gy in weeks 5 -6.

Dose accumulation: adaptation versus non-adaptation
For both scenarios, i.e. adapted and non-adapted, signifi cant differences were found compared to the initial plan. Patient averaged D 2% and D mean doses for all OARs and averaged D 2% and D 98% for three target volumes are summarized in Table II For the spinal cord and brainstem the dosimetric differences between the adaptive and nonadaptive approach were generally small for VMAT, while they were statistically signifi cant for D 2% doses in case of IMPT. In the adapted scenario median D mean for parotid glands and the larynx were also lower, but not statistically signifi cant for neither of the modalities. The D 2% difference for target volumes between adapted and non-adapted strategies was found to be statistically signifi cant for both modalities, e.g. for PTV 70Gy it decreased by more than 4.4 Gy for IMPT. D 98% for VMAT did not vary much between the adapted and non-adapted treatment concepts, except for PTV 70Gy where the median D 98% decreased after adaptation. For IMPT D 98% values increased in general after plan adaptation, however without statistical signifi cance.

Discussion
The study indicated the importance of repeated imaging in fractionated radiotherapy for H & N cancer. Biweekly dose evaluation was considered to be adequate to quantify the difference in dosimetric parameters for target structures and OARs. When applying the initial plan on the repeated CT obtained at week 2 and 4, increased OAR doses were found paired with deteriorated target dose conformity. Nevertheless, in half of the patients, treatment plan parameters were still acceptable with VMAT (V 95% Ͼ 95%, OARs within limits) after recalculating the initial plan on the updated CTs. Treatment plan adaptation, performed for those patients, optimized OARs sparing even further. For IMPT none of the recalculated plans was found to be acceptable in terms of target coverage, since even small anatomic variations had a large impact on the resulting dose distributions.
A 5 mm expansion from each CTV was applied to construct PTVs, for both modalities, which is taken from our photon experience. For proton therapy there are different concepts for PTV construction, ranging from beam specifi c margins, anisotropic margins or more sophisticated methods including robust optimization. However, there are proton facilities which use 5 mm margin expansion for H & N cancer patients in clinical routine [6,15]. Finding the best margin strategy was outside the scope of the  study, because we intended to compare directly proton plans with VMAT plans under similar conditions, i.e. we wanted to evaluate the magnitude of dose degradation for the same volume changes.
Even though the magnitude of target coverage changes was large for proton plans, OAR dose limits were never exceeded, even not in the non-adapted approach. This can be explained by the superior initial OARs sparing for IMPT. For VMAT plans OAR doses in the initial plans were very often close to planning constraints and therefore even a small dose increase violated the planning aims. Consequently, it is not straightforward to translate adaptive approaches from the photon experience towards proton therapy. Therefore, it is essential to explore ART for protons separately.
This study was based on two rescans in two weekly intervals. Repetitive CTs on a daily/weekly basis could provide even more information about anatomical variations. At present there is no consensus on how many image series needed to be taken during the course of treatment or how often replanning should be executed. Obviously, if repeated imaging is utilized for ART, the extra dose coming from additional imaging, cannot be neglected [22]. Wu et al. considered different replanning schemes for IMRT and concluded that replanning more often than once a week is unnecessary [7]. However, this might not be valid for IMPT.
The deformation of the CT images was performed in the anti-chronological way, which means that CT 1 and CT 2 were registered to CT 0 and the accumulated dose was projected to CT 0 . Chronological image deformation and dose accumulation onto the last CT would bring slightly different results due to the difference between the deformed structure and the actual structure drawn by the radiation oncologist on repeated CTs. However, if delineated and deformed structures would be identical no difference should be found between the chronological and anti-chronological dose summation. Nevertheless, using deformed structures without inspection and adaptation by an experienced radiation oncologist is not possible, as current state-of-the-art DIR algorithms are not robust enough. It is still under discussion which summation method should be used [15,23].
Although conformal VMAT and IMPT dose distribution were created at the time of treatment planning based on CT 0 , recalculating a plan on the repeated CTs (CT 1 or CT 2 ) with a different, i.e. smaller, target volume, resulted in a less conformal dose distribution, but with suffi cient coverage. For the ART approach, new plans, designed especially for the smaller target, were highly conformal and with high coverage on each repeated CT. After DIR and dose accumulation the smaller target did not exactly reach the initial target size, therefore after dose summation the target coverage (for some cases) was lower compared to non-adapted plans. This is also refl ected in decreased D 98% values for the target structures. This imperfect behavior of current DIR algorithms with respect to changing tumor volume is a general limitation of ART. For OARs volume changes are less pronounced than for target structures, and consequently assessing the total OAR dose picture in ART by dose accumulation deems to be warranted.
In summary, the assessment of the dosimetric consequences of anatomical variations underlines the benefi ts of adaptive strategies for both photon and proton therapy. However, the advantages of ART were largely patient dependent. Moreover, the same anatomical variations resulted in a very different impact on dose distributions for IMPT and VMAT. Therefore, the clinical implementation of ART needs to be adjusted to needs of the underlying treatment modality.