Effect of intrafraction adaptation on PTV margins for MRI guided online adaptive radiotherapy for rectal cancer

Purpose To determine PTV margins for intrafraction motion in MRI-guided online adaptive radiotherapy for rectal cancer and the potential benefit of performing a 2nd adaptation prior to irradiation. Methods Thirty patients with rectal cancer received radiotherapy on a 1.5 T MR-Linac. On T2-weighted images for adaptation (MRIadapt), verification prior to (MRIver) and after irradiation (MRIpost) of 5 treatment fractions per patient, the primary tumor GTV (GTVprim) and mesorectum CTV (CTVmeso) were delineated. The structures on MRIadapt were expanded to corresponding PTVs. We determined the required expansion margins such that on average over 5 fractions, 98% of CTVmeso and 95% of GTVprim on MRIpost was covered in 90% of the patients. Furthermore, we studied the benefit of an additional adaptation, just prior to irradiation, by evaluating the coverage between the structures on MRIver and MRIpost. A threshold to assess the need for a secondary adaptation was determined by considering the overlap between MRIadapt and MRIver. Results PTV margins for intrafraction motion without 2nd adaptation were 6.4 mm in the anterior direction and 4.0 mm in all other directions for CTVmeso and 5.0 mm isotropically for GTVprim. A 2nd adaptation, applied for all fractions where the motion between MRIadapt and MRIver exceeded 1 mm (36% of the fractions) would result in a reduction of the PTVmeso margin to 3.2 mm/2.0 mm. For PTVprim a margin reduction to 3.5 mm is feasible when a 2nd adaptation is performed in fractions where the motion exceeded 4 mm (17% of the fractions). Conclusion We studied the potential benefit of intrafraction motion monitoring and a 2nd adaptation to reduce PTV margins in online adaptive MRIgRT in rectal cancer. Performing 2nd adaptations immediately after online replanning when motion exceeded 1 mm and 4 mm for CTVmeso and GTVprim respectively, could result in a 30–50% margin reduction with limited reduction of dose to the bowel.


Introduction
Neo-adjuvant (chemo)radiotherapy plays an important role in the multidisciplinary treatment of rectal cancer [1], primarily aiming to reduce local recurrence rates [2,3] and to downstage the tumor prior to surgery. Accurate radiotherapy delivery to the tumor and elective lymph nodes is hampered by geometrical uncertainties arising from delineation uncertainty, and inter-and intrafraction anatomical variations. To accommodate these uncertainties, the clinical target volume (CTV) is expanded to a planning target volume (PTV). This target volume typically overlaps with the organs at risk (OAR) such as the bladder and the small bowel, resulting in high OAR dose and consequent toxicity [4,5]. Studying mesorectum motion is important for optimizing rectal cancer Open Access *Correspondence: u.vd.heide@nki.nl Department of Radiation Oncology, The Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands radiotherapy in which the mesorectum receives a homogeneous dose in either short or long treatment schedules. Within the context of organ preservation for intermediate and high risk rectal cancer patients, safe dose escalation to the primary tumor may be enabled with the use of smaller PTV margins around the GTV [6].
Recently, integrated MRI linear accelerators were introduced, allowing the use of MRI for online image guidance. With MRI-guided radiotherapy (MRIgRT) high soft tissue contrast images can be acquired at several time points during the treatment which enables daily online adaptation to anatomical changes between treatment fractions and monitoring of anatomical changes during treatment [7]. By planning using delineations of the anatomy on images acquired just prior to the treatment, MRIgRT allows for reduction of geometrical uncertainties due to interfraction motion. As a result of daily online adaptation, intrafraction motion and delineation uncertainty are the primary remaining uncertainties [8].
Online adaptation for rectal cancer is time-consuming with a median duration of 36 min [9] as it requires online redelineation and plan optimization based on the image of the day. As demonstrated by Kleijnen et al. [10], intrafraction motion increases with time requiring larger PTV margins for longer treatment durations. Ideally, to reduce intrafraction motion, online adaptation could be accelerated by automated methods, like auto-contouring or auto-planning, however these methods are still in development for routine clinical use [11].
Strategies for intrafraction motion monitoring and subsequent motion management, including beam gating and multi-leaf collimator tracking, allow for the reduction of uncertainties arising from intrafraction motion. With gating, the target position is monitored continuously and radiation is only delivered if the target is within a pre-defined envelope. Gating has been widely implemented, but its application is mostly limited to periodic motion [12,13]. Although rectal motion is non-periodic, gating has been applied for mitigating rectal intrafraction motion [14]. Next to gating, tracking has been investigated [15,16], although it is not clinically available to date on MRI linear accelerators. An alternative, simpler, intrafraction adaptation strategy is to acquire a verification MRI to evaluate target motion during redelineation and plan adaptation, and perform a 2nd adaptation if the target has moved outside a pre-defined envelope. This adaptation can be done by repeating the workflow for the initial adaptation. Adapting based on the verification MRI will probably provide a better surrogate for the anatomy on the post treatment MRI than the adaptation MRI considering the shorter time interval between the scans. As a result of the shorter time interval, target motion may possibly be smaller and a further reduction of PTV margins may be possible. Margin reduction has been studied for prostate [17,18], lung [19], cervical [20,21] and spine irradiation [22], however no studies focusing on the potential benefit of intrafraction motion management on PTV margins for rectal cancer were found.
The aim of this work was therefore to determine the PTV margins required to accommodate intrafraction motion of the mesorectum during standard MRIgRT and of the primary tumor during dose-escalated MRIgRT of rectal cancer and secondly to determine the potential benefit of performing a 2nd adaptation prior to irradiation.

Patient data
Data of 30 patients with intermediate risk or locally advanced rectal cancer treated on a 1.5 T MR-Linac (Unity, Elekta AB, Stockholm Sweden) between October 2018 and March 2021 were analyzed. Twentytwo patients received short course radiotherapy (SCRT; 5 × 5 Gy) and 8 received long course chemoradiotherapy (LCRT; 25 × 2 Gy). Ethics approval was obtained and all patients provided written informed consent for use of their data. Patients were treated using an online adaptive workflow (Fig. 1A) [23]. Approximately a week before start of the treatment, a planning CT and MRI were acquired on which the elective target volumes and organs at risk (OAR) were delineated for treatment planning according to delineation guidelines [24,25]. For SCRT patients the CTV consisted of the mesorectum, the presacral region and pelvic lymph node regions, while for LCRT patients the lymph node region included the obturator region if pathological lymph nodes in situ were identified. The PTV was generated by adding an anterior anisotropic margin for the mesorectum CTV and a uniform 5 mm margin for the lymph node regions as suggested by Valentini et al. [24]. A bladder filling protocol consisting of drinking 250 ml water 30 min prior to simulation and the radiotherapy session on the MR-Linac was advised. On the MR-Linac, during each fraction first a 3D-T2-weighted MRI was acquired for adaptation (MRI adapt ) and the planning CT was deformably registered to MRI adapt after which the elective target volumes were re-delineated for SCRT patients. Based on these new delineations, a new plan was optimized. For LCRT the plan corresponding to the actual mesorectum shape was selected through the use of a library of plans [26], followed by a virtual couch shift to correct for set-up errors [23]. Prior to starting the irradiation, an additional MRI was acquired to verify the position of the structures (MRI ver ). Subsequently, all patients were irradiated with an optimized 9-field intensity modulated radiotherapy (IMRT) plan with beam avoidance angles for the cryostat pipe (gantry angles 8°-18°) and two high attenuation regions of the MRL treatment couch (100°-140° and 220°-260°) [27]. After irradiation, a post treatment MRI (MRI post ) was acquired.
All available images acquired during 5 daily treatment fractions of the patients treated in a short course scheme and the first fractions of every week of patients treated in a long course scheme were used for this study. The time intervals between MRI adapt and MRI ver , and MRI ver and MRI post were determined. These intervals correspond to the time needed for recontouring and plan adaptation, and irradiation respectively. On all the images, the gross tumor volume of the primary tumor (GTV prim ) and the mesorectum clinical target volume (CTV meso ) were delineated retrospectively using the contouring toolbox in Monaco v5.40.01 (Elekta, Stockholm, Sweden) by 2 experienced radiation technology therapists (RTT) following delineation guidelines [24,25]. For each fraction, delineations of MRI adapt were copied to the MRI ver and MRI post , and manually adjusted. All scans of one patient were delineated by the same RTT. Delineations were verified and, if needed, corrected by a radiation oncologist with over 10 years' experience. We indicate the mesorectum CTV as delineated on the MRI adapt with CTV meso,adapt in the remainder of this paper. We use a similar convention for the other scans.
The peritoneal cavity (bowel area) as delineated on MRI adapt of the first fraction was used and adjusted if needed. The CTV of the elective lymph node regions was not included in this study, considering the intrafraction motion of these regions is expected to be small [9]. For the same reason, a 2nd adaptation is suspected to have no substantial effect.

PTV margin determination
For every fraction, the delineated structures (GTV prim,adapt and CTV meso,adapt ) were expanded in 3D to new structures: PTV prim and PTV meso in steps of 1.0 mm. The expansions were obtained using a rolling-ball algorithm [28], where the expansions were simulated on the actual scan. For GTV prim,adapt the margin was isotropic and for CTV meso,adapt we used anisotropic margins with the anterior expansion 1.6 times larger compared to all other directions. This choice is motivated by the study of Nijkamp et al. on mesorectum shape variation [29]. For every expansion the coverage was determined between the PTV and the associated structure on MRI post . CTV meso and GTV prim were analyzed separately and Steps not relevant to current study are not included in the workflow independently. The coverage was defined as the number of overlapping voxels of the PTV and corresponding structures on MRI post as a percentage of the total number of voxels of the structure on MRI post . For CTV meso , we considered the margin adequate when on average over the 5 fractions, PTV meso covered 98% of CTV meso,post in 90% of the patients. For the boost to the GTV prim the criterion was relaxed to 95% volumetric coverage of GTV prim,post by PTV prim in 90% of the patients. The reason for using a lower coverage criterion is because the GTV prim would be irradiated as a boost on top of the irradiation of the CTV meso , resulting in less steep dose gradients. Considering the heuristic choice of these coverage criteria, we also assessed the effect of different volumetric coverage criteria on the PTV margins. All expansion and coverages were calculated using in-house software Match42.

Second adaptation
To study the effect of a 2nd adaptation after MRI ver , the intrafraction motion during redelineation and plan adaptation was evaluated. The verification envelope (VE) is defined as an expansion around GTV prim,adapt or CTV meso,adapt . We consider the threshold for intrafraction motion during adaptation to be exceeded, when the VE does not cover the GTV prim,ver or CTV meso,ver for at least 95% and 98% respectively.
For an isotropic VE, we determined whether the intrafraction motion during redelineation and plan adaptation exceeded the threshold for GTV prim . For CTV meso the VE was anisotropic with a similar anterior expansion a factor 1.6 times all other directions. If the threshold was not exceeded, no 2nd adaptation would be needed and thus GTV prim,adapt or CTV meso,adapt was used for evaluation. If the threshold was exceeded, a 2nd adaptation was applied and GTV prim,ver and CTV meso,ver were used to generate a new PTV prim and PTV meso to represent this adaptation. The online adaptive workflow following this approach is shown in Fig. 1B.
We varied the VE between 0 and 10 mm with 1 mm increments and we determined for each size the required PTV margin and the frequency of 2nd adaptations needed. This process is summarized in Fig. 2.
To determine the potential benefit of a 2nd adaptation for organs at risk, we determined the overlap between the bowel area and the required PTV meso when no 2nd adaptations are performed and compared this to the overlap between the bowel area and PTV meso after 2nd adaptations.
As a surrogate for the time required for re-delineation during the first and 2nd adaptation, we consider the change in volume of the CTV meso. This is a practical surrogate, since re-delineation of CTV meso is in particular necessary due to changes in rectal filling, which can directly influence the volume. For the first adaptation For each value we first determine whether to perform a 2nd adaptation and next we determine the required margin we consider the volume change between the reference CT and MRI adapt . For the 2nd adaptation we consider the change between MRI adapt and MRI ver. A paired sample t-test was carried out (in SPSS v27.0) to test for a significant difference (α = 0.05).

Results
For 4 out of 30 patients one or more MRI post were not available and one patient received a treatment fraction on the conventional linear accelerator resulting in a total of 144 fractions available for analysis. For one patient the GTV prim was poorly visible on MRI and therefore not delineated, resulting in a total of 139 fractions for analysis of GTV prim . Patient and tumor characteristics are summarized in Table 1.

PTV margins and 2nd adaptations
The median time between MRI adapt and MRI ver was 12 min (inter-quartile range IQR = 10-23 min) and between MRI ver and MRI post 12 min (IQR = 11-15 min). The margin required for PTV meso without 2nd adaptations was 6.4 mm in the anterior direction and 4.0 mm in all other directions. This is indicated in Fig. 3 for a VE of 10 mm, in which case no 2nd adaptations are needed. For PTV prim a margin of 5.0 mm was required.
The coverage for all patients (n = 30) is shown in Fig. 4. In 90% of the population, the target criteria of 98% and 95% coverage were reached. The coverage of the remaining 10% was somewhat lower, but still above 88% in all cases.
The percentage of 2nd adaptations and resulting PTV margins are shown in Fig. 3. The minimal feasible margin for PTV meso when performing 2nd adaptations was 3.2 mm in the anterior direction and 2.0 mm in all other directions using a VE of 1.0 mm. To achieve this margin, 2nd adaptations needed to be performed in 36% of the fractions.. Adapting more fractions did not lead to further margin reduction. For GTV prim , 2nd adaptations would be needed in 17% of the fractions at a VE of 4.0 mm to reduce PTV prim to 3.5 mm. Table 2 shows the number of patients needing a 2nd adaptation in 0, 1,2,3 of 4 fractions. For CTV meso 7 patients needed a 2nd adaptation in 3 or more fractions. One of these patients was treated with LCRT. For GTV prim , only 2 patients with   Table 3. For the reduced PTV meso that was possible after 2nd adaptations, 19.3 (12.6-34.7) cm 3 of the bowel overlapped with PTV meso .
The CTV meso showed larger changes between the planning CT and MRI adapt than between MRI adapt and MRI ver , as reflected by a median volume difference (IQR) of 26.1 (14.6-45.7) vs 7.2 (4.8-12.9) cm 3 ; p < 0.05.

Effect of coverage criteria on PTV margins
For different coverage criteria, PTV margins for a workflow without 2nd adaptations (blue line) and when the minimum feasible margin is reached when performing 2nd adaptations (red line) are shown in Fig. 5. For both scenarios the margin increases gradually as more volumetric coverage is required. Above 97% a steeper increase of the margins is seen for PTV meso as compared to the lower coverage criteria.

Discussion
The aim of this study was to determine the PTV margins required to accommodate intrafraction motion of the mesorectum (CTV meso ) and the gross tumor volume (GTV prim ) during MRIgRT of rectal cancer and to determine if performing a 2nd adaptation prior to irradiation would potentially be beneficial.
For the CTV meso we found a required margin of 6.4 mm in the anterior direction and 4.0 mm in all other directions without a 2nd adaptation. Introducing 2nd adaptations allowed a reduction to 3.2 mm in the anterior direction and 2.0 mm in all other directions. For the GTV prim , a PTV margin of 5.0 mm was needed, whereas 2nd adaptations allowed for a reduction to 3.5 mm.
Kleijnen et al. studied the motion uncertainty as a function of time of CTV meso and GTV prim using repeated cine-MRI data of 16 patients [10]. They found PTV margins of 12 mm for intrafraction motion up to 18 min which were comparable in magnitude to margins found for interfraction motion [10]. The differences are likely due to the use of different coverage criteria. In the study of Kleijnen et al., the distance that incorporates 95% of the surface voxels at the investigated time point was required to fit within the margin in 90% of all fractions. In our work the margin was selected for an average volumetric coverage of 95% in 90% of all patients.
With regards to PTV prim, our findings are in line with previous studies [33,34]. Van de Ende et al. studied the inter-and intrafraction displacement of the GTV based on fiducial markers on cone beam CT images and reported PTV margins of 3.0 mm in left-right direction, 4.7 mm in anterior-posterior direction and 5.5 mm in cranial-caudal direction for intrafraction displacement [33]. In addition, they showed larger motion for proximal tumors as compared to distal tumors and hypothesized that the reduction of required margins may be higher in patients with a proximal compared to a distal tumor.
More recently, Eijkelenkamp et al. determined margins to compensate for intrafraction GTV prim motion during online adaptive procedures [34]. They used a similar method as the current study to determine the required margin for online adaptive MR-guided dose escalation for intermediate risk rectal cancer patients and reported a margin of 6 mm for the entire treatment, which could be reduced to 4 mm for a procedure of 15 min or less. These findings are consistent with the PTV margins found in current study.
Although intrafraction motion for CTV meso and GTV prim has been studied previously, the current study also explores the potential benefit of intrafraction motion management during MRI-guided radiotherapy to reduce the required PTV margins. As shown in the results, adapting just prior to the start of irradiation instead of only at the beginning of the treatment possibly provides a more accurate estimation of the anatomy during irradiation in some cases, given the shorter time interval between MRI ver and MRI post compared to MRI adapt and MRI post . For GTV prim relatively more 2nd adaptations were needed to achieve a margin reduction of 30%. This may possibly be attributed to the larger observer variability for the primary tumor as compared to mesorectum [35]. In addition, when considering the number of 2nd adaptations and the resulting margins for different verification envelopes as depicted in Fig. 3, one can make a tradeoff between the workload and the benefit of motion management to reduce the required margins.
When assessing the effect of the margin reduction on bowel toxicity, we showed that the volume of the bowel receiving 95% of the prescribed dose (23 Gy) is reduced with only 16.0 cm 3 after performing 2nd adaptations. Both before and after 2nd adaptations, the volume of bowel area receiving 23 Gy was lower than the upper limit of 85 cm 3 recommended by adapted Quantitative Analysis of Normal Tissue Effects in the Clinic (QUAN-TEC) guidelines [36]. Considering this, the clinical impact of margin reduction as a result of 2nd adaptations might be limited for dose reduction to the bowel. Overall the choice to treat a patient on a MR-Linac systems is carefully considered by clinicians weighing the clinical benefits against the added time and workload. With daily online adaptation and motion management where needed the treatment is tailored to each patients' anatomy, allowing for more accurate RT, reduced margins and possibly dose escalation.
In this study we introduced a verification envelope (VE) for deciding when to perform a 2nd adaptation. The coverage threshold based on this VE replaces the common practice to take action if the target moves out of the PTV. As demonstrated in the results, the required PTV margin is typically not identical to the VE. Using the PTV margin as envelope can result in performing either too few or too many 2nd adaptations than necessary. The concept of a VE is consistent with motion management techniques such as automated beam gating [37] and target tracking [38] as these are solely based on movement of the target outside of a pre-specified threshold, and do not inherently use the PTV margin for this purpose. Up to date there is one study by Chiloiro et al. [14] demonstrating the clinical feasibility of beam gating in rectal cancer with a region of interest set around the mesorectum.
A limitation of this study was that the duration of the 2nd adaptations was not considered. We assumed an instantaneous adaptation, which is not feasible in clinical practice. When performing the 2nd adaptation, motion may occur during that time period as well. Consequently, the margins found in this study should be considered a lower boundary of what would be achievable. Our method for 2nd adaptations uses new delineations on MRI ver , which would be consistent with an Adapt to Shape workflow on the Unity system. Although full redelineation as part of a 2nd adaptation is the most accurate approach to account for intrafractional anatomical changes, this method tends to be time-intensive, but is expected to be faster than the first adaptation. Because we assume the 2nd adaptation to be faster, we used the volume differences between the contours prior to and after adaptation as an estimate for the adaptation time.
Here volume differences were used as a surrogate for the added path length [39] and we saw that volume differences were significantly smaller for the 2nd adaptation.
At the time of adapting for the 2nd time, the patient has been on the treatment table for a while and may be more relaxed, possibly resulting in a reduced amount of motion as compared to the first adaptation. An option to limit the adaptation time might be to opt for a less accurate and faster approach such as Adapt to Position [23]. However, a downside is that the Adapt to Position approach only corrects rigid translations of the target volume. Nevertheless, the exact implications of the duration of 2nd adaptation remain to be studied further. Speeding up the first adaptation, specifically delineation, might be the ideal solution. However, automation methods such as auto contouring are still in development.
The criteria for margin determination were based on volumetric coverage and not statistical inferences from accumulated dose as is done in deriving the classical margin recipes [40,41]. For a comparison with these recipes the local standard deviation of the positioning error should have been determined. However, to translate this into a margin, assumptions on the dose distribution, local distribution of positioning errors and target deformation have to be made. Our volumetric approach is considerably easier to interpret and requires only the choice of a coverage criterion. However to formally asses that, a dose accumulation study needs to be performed. Because of the heuristic nature of these choices we also provided results for different coverage criteria.
The proposed margins primarily account for uncertainties due to intrafraction motion conform the online adaptive workflow. In the total PTV used in clinical practice other uncertainties such as uncertainties in gantry positioning, MLC motion, image alignment should be included. Gantry position and choices related to MLC positioning are typically institute-specific. When using this work to determine margins for clinical practice, care should be taken to ensure that all relevant uncertainties are taken into account.
Given the comparison of two delineations on different scans, the analysis is potentially influenced by delineation variability. However, within a single patient we minimized this variation by having the same observer delineate all scans of one patient. Moreover, for the verification and post treatment scan the delineation was performed by adjusting a copy of the delineation on the adaptation scan, minimizing the delineation variability within a single fraction.

Conclusion
Our study shows that the PTV margins to accommodate intrafraction motion of the mesorectum in online adaptive MRIgRT for rectal cancer are 6.4 mm in anterior direction and 4.0 mm in other directions, and 5.0 mm for the GTV of the primary tumor. In this study we introduced a verification envelope based on which the decision is made on when to perform a 2nd adaptation. Even in the most optimistic scenario motion management in the form of a 2nd adaptation prior to irradiation, these margins can be reduced to 3.2 mm in anterior direction and 2.0 mm in other directions for the mesorectum and 3.5 mm for the primary tumor, with limited reduction of dose to the bowel.