Impact of abdominal compression on intra-fractional motion and delivered dose in magnetic resonance image-guided adaptive radiation ablation of adrenal gland metastases

Purpose: The current study investigated the impact of abdominal compression on motion and the delivered dose during non-gated, magnetic resonance image (MRI)-guided radiation ablation of adrenal gland metastases. Methods: Thirty-one patients with adrenal gland metastases treated to 45 – 60 Gy in 3 – 8 fractions on a 1.5 T MRI-linac were included in the study. The patients were breathing freely (n = 14) or with motion restricted by using an abdominal compression belt (n = 17). The time-resolved position of the target in online 2D cine MR images acquired during treatment was assessed and used to estimate the dose delivered to the GTV and abutting luminal organs at risk (OAR). Results: The median (range) 3D root-mean-square target position error was significantly higher in patients treated without a compression belt [2.9 (1.9 – 5.6) mm] compared to patients using the belt [2.1 (1.2 – 3.5) mm] (P < 0.01). The median (range) GTV V95% was significantly reduced from planned 98.6 (65.9 – 100) % to delivered 96.5 (64.5 – 99.9) % due to motion (P < 0.01). Most prominent dose reductions were found in patients showing either large target drift or respiration motion and were mainly treated without abdominal compression. Motion did not lead to an increased number of constraint violations for luminal OAR. Conclusions: Acceptable target coverage and dose to OAR was observed in the vast majority of patients despite intra-fractional motion during adaptive MRI-guided radiation ablation. The use of abdominal compression significantly reduced the target position error and prevented the most prominent target coverage degradations and is, therefore, recommended as motion management at MRI-linacs.


Introduction
The adrenal glands are common sites for metastatic spread from various tumours [1].While surgery is often the preferred treatment choice, many patients with adrenal gland metastases are not candidates for adrenalectomy due to poor performance status or more than one metastatic site [2].As an alternative, stereotactic body radiotherapy (SBRT) using a high biological equivalent dose (BED) and steep dose gradients is a safe and effective treatment [3,4].Based on multiinstitutional data, recent models of tumour control probability in SBRT for adrenal metastases show that a target dose of a BED of 100 Gy or more is needed to achieve high local control [5,6].However, due to the mobility and radiosensitivity of the organs at risk (OAR), SBRT of targets in the abdomen is technically challenging.Delivering such a high dose near radiosensitive luminal organs poses a risk of inducing severe adverse events.This is especially the case when treatment is performed on conventional cone-beam computed tomography (CBCT) imageguided linear accelerators, where the limited soft-tissue contrast challenges the localisation of the target and OARs.
In recent years, magnetic resonance (MR) image-guiding treatment systems such as MR linear accelerators (MRI-linac) have been introduced in a clinical setting [7][8][9][10][11][12][13]. The MR image guidance provides superior soft-tissue contrast and allows for online plan adaptation addressing the inter-fractional volumetric changes in the target and OARs.Some MR image-guiding systems provide real-time gated treatment delivery based on soft-tissue structures in 2D cine MR images [7,14].Previous studies have reported using such treatment-gating solutions to manage respiration motion during MR image-guided SBRT of adrenal gland metastases [15][16][17].However, treatment gating is still not clinically implemented on all MR image-guiding systems.Furthermore, gating leads to prolonged treatment time [18] and hence a risk of reduced compliance and radiobiological effect.Therefore, gating might not be preferred if non-gated solutions give the target an adequate dose without compromising the OAR constraints.An alternative to gating is reducing respiratory motion using abdominal compression techniques [19].Such motion management techniques have previously been used for non-gated MR image-guided SBRT of pancreatic cancer [20,21].
The current study aimed to retrospectively investigate the impact of abdominal compression on intra-fractional motion and delivered target and abutting OAR dose in a cohort of patients with adrenal gland metastases treated using non-gated MR image-guided adaptive SBRT.

Patient population
Until July 2022, thirty-two patients with metastasis in the adrenal gland underwent MR image-guided adaptive radiotherapy at our hospital.All patients provided informed consent for registry in the prospective Multi-OutcoMe EvaluatioN of radiation Therapy Using the MRlinac (MOMENTUM) study (NCT04075305) [22].One patient was not included in the current study due to missing 4D computed tomography (CT) data, which is needed to extract peak-to-peak target respiration motion as described below.The relevant patient characteristics for the remaining thirty-one patients included in the study are summarised in Table 1.

Imaging and treatment protocol
Depending on capabilities, the patients were immobilised with both arms up (n = 8), one arm up (n = 17), or both arms down (n = 6).From May 2021, immobilisation included using an abdominal compression Velcro belt (Danish Healthcare ApS, Ballerup, Denmark) to assist the patients in breathing steadily and reducing respiratory-induced target motion.The belt was adjusted at the planning scan to maximal tolerable tightness, and the belt was marked with a pen indicating a patientspecific circumference to ensure reproducibility during treatment.All patients except one with an ostomy tolerated the use of the belt.
Pre-treatment MR simulation was performed with the patient in treatment position on a 1.5 T Philips Ingenia MRI scanner (Philips Medical Systems BV, Best, The Netherlands) with a 3D T2w image (1400 ms repetition time, 137 ms echo time, 90 • flip angle, 448 × 448 matrix, 1 × 1 mm 2 pixel size, 2 mm slice thickness, 25 cm scan length).During imaging, the patients were breathing freely (n = 14) or had breathing motion restricted by using the abdominal compression belt (n = 17).Information about electron densities and baseline respiration-induced target motion was obtained from a 4D-CT scan (512 × 512 matrix, 1 × 1 mm 2 pixel size, 3 mm slice thinkness) acquired on either a Toshiba Aquillon On (Canon Medical Systems Corporation, Otawara, Japan), a Philips Big Bore Brilliance CT (Philips Medical Systems BV, Best, The Netherlands), or a Siemens Somatom go.Now (Siemens Healthcare GmbH, Erlangen, Germany).Electron density information in bulk regions was transferred using deformable image registration to the MR reference image for dose planning [12].
Pre-treatment target motion information was extracted from 4D-CT using a customised workflow in MIM Maestro (MIM Software Inc., Beachwood, OH, USA) as described previously [23].The peak-to-peak distance of the target was obtained in the left-right (LR), anteriorposterior (AP), and inferior-superior (IS) directions.Patient-specific, anisotropic GTV-PTV margins depending on the target peak-to-peak respiration motion extent in each direction were employed (see Tables 1 and 2).
Patients were prescribed 45 Gy in three fractions (n = 12), 50 Gy in five fractions (n = 17), or 60 Gy in eight fractions (n = 2) depending on the target size and proximity of critical OAR at the discretion of the treating physician.The gross tumour volume (GTV) should preferentially be covered by the 95% isodose line (GTV V95% >99%), and the mean GTV dose should be >100% (GTV D mean ≥ 100%).The planning target volume (PTV) should be covered by the 67% isodose line (PTV V67% >99%).Compromise of the target coverage was enforced to meet hard dose constraints for OAR (see Table 3).The distance from the GTV to the closest luminal OAR (duodenum, oesophagus, large bowel, small bowel, or stomach) was 0 mm (n = 9), 1-5 mm (n = 9), 6-10 mm (n = 3), or above 10 mm (n = 10).Reference dose plans were created in Offline Monaco (Elekta AB, Stockholm, Sweden) using 11 beams, excluding gantry angles where the beam would pass through the patient's arms, the cryostat pipe, or highdensity structures in the treatment couch edges.A maximum of 50-60 segments, a minimum segment area of 2-4 cm 2 and a minimum of 5 Monitor Units per segment were used.The statistical uncertainty of the Monte Carlo dose engine used for dose calculation was set to 1% per calculation, and the dose grid size was 3 mm.A sample pre-treatment MR scan and reference dose plan are shown in Fig. 1.
MR image-guided adaptive radiotherapy using an adapt to shape workflow (re-delineation of target and OAR followed by plan adaptation) [24] was carried out on the Elekta Unity 1.5 T MRI-linac (Elekta AB, Stockholm, Sweden).At each treatment fraction, the patient was positioned on the MRI-linac table, and a session 3D T2w scan (1400 ms repetition time, 137 ms echo time, 90 • flip angle, 408 × 408 matrix, 1 × 1 mm 2 pixel size, 2 mm slice thickness, 25 cm scan length) was acquired.During both imaging and treatment, the patients were breathing freely or had breathing motion restricted by using the abdominal compression belt.Daily adapt to shape workflow was used where the GTV, OAR, and density contours were propagated from the reference scan to the session scan and manually edited at the discretion of the present physician.A 3D T2w, position verification scan was acquired during plan adaptation to ensure the target had not moved.Before beam on, motion monitoring consisting of 2D-slice interleaved coronal and sagittal cine MR images (3.8 ms repetition time, 1.9 ms echo time, 40 • flip angle, 352 × 352 matrix, 1.2 × 1.2 mm 2 pixel size, 5 mm slice thickness) acquired at 2.5 Hz for each plane was initiated.The acquired image planes intersect at the centre of the bounding box of the PTV [25].Motion monitoring was switched off before beam-off to allow for the acquisition of a third 3D T2w scan, which ended approximately simultaneously with the end of the dose delivery.

Intra-fraction target motion analysis
Three-dimensional target motion profiles were extracted from the motion monitoring cine MR images in Matlab (Mathworks Inc., Natick, MA, USA) using an in-house tracking algorithm described in detail in the supplementary materials (see Supplementary Fig. S1).Validation of the tracking algorithm using a 4D motion phantom is also described in the supplemental materials (see Supplementary Table S1 and Supplementary Fig. S2).An example of the visual output of the tracking algorithm is shown in the animated Supplementary Fig. S3.Occasionally, motion monitoring malfunctioned or was switched off prematurely, and treatment fractions with less than three minutes of cine imaging data were excluded from the study.This time frame was chosen to allow for the monitoring of at least part of the potential drift occurring during treatment.A typical beam-on time on the 1.5 T MRI-linac is approximately one minute per Gy of the prescribed dose [12].
The total 3D root-mean-square position error (mean target position error during treatment) was determined from the motion profiles at each fraction and averaged over all fractions as an overall measure of the patient-specific target offset error.Drift was disentangled from breathing motion by applying a moving average filter (see example in Fig. 2).Breathing motion was estimated by subtracting the drift from the raw displacement values.At a given fraction f, peak-to-peak respiration motion amplitudes in each of the principal directions, PP LR,f ,PP AP,f , and PP IS,f , were estimated by the difference between the minimum and the maximum breathing displacements, excluding the top and bottom 2.5 percentiles to reduce the effect of outliers [21].The overall target peakto-peak respiration motion at a given fraction was estimated by the root of the sum of squares of distances in each direction The total online target peak-to-peak respiration motion for each patient was found by averaging PP f values over all fractions.The mean target drift in each of the principal directions at a given fraction, D LR,f , D AP,f , and D IS,f were given by the absolute value of the average drift of the fraction.The overall target drift at a given fraction was estimated by the root of the sum of squares of values in each direction The total online target drift motion for each patient was found by averaging D f values over all fractions.

Delivered dose estimation
The effect of the intra-fraction target movements during beam delivery on the dose delivered to the GTV and the luminal OAR closest to the GTV was assessed by shifting the structures in space according to the displacements found by the tracking algorithm while doing dose summation from the static dose grid during the process.Thus, doses were added to each voxel of the structures after each shift without performing any dose recalculation.Assessing the total summed dose to the structure of a given patient during the treatment course was accomplished through a simple summation of the dose volume histogram (DVH) from each treatment fraction.The consequences of motion were evaluated by a population median GTV DVH, by changes in GTV V95% and mean GTV

Table 2
Calculation of the PTV margins.The calculation was based on the peak-to-peak target motion assessed on 4DCT in each direction.The target motion was evaluated in the planning images and was used without any additional correction related to the compression belt.If a compression belt was used, it was used both during the treatment planning scan and at all treatment fractions.*The PTV margin for one patient deviated from the scheme as a PTV margin of 6 mm was used despite a peak-to-peak target motion of 17.6 mm.dose, as well as by OAR D0.1 cc and D1cc, for which there were hard constraints for the luminal organs.

Statistical methods
Non-parametric tests were employed.Statistical differences between groups and within groups were tested using Wilcoxon rank sum tests, and Wilcoxon signed rank tests, respectively.Association between variables was investigated using Spearman's rank correlation coefficient.For the population median DVH, a p-value was calculated per dose level to indicate larger regions with p-values below 5%, as conceptualised in Bertelsen et al. [26].The calculation of population median values is based on one value (the mean over all fractions) per patient to account for the varying number of treatment fractions received per patient.

Results
All 31 patients completed treatment, and all 137 fractions were delivered.One treatment fraction was interrupted during beam-on, and patient setup was repeated before the rest could be delivered.Cine data from that treatment fraction was excluded from the analysis.Due to software malfunctioning or user mistakes, motion monitoring was not working in ten fractions divided among nine patients.Furthermore, insufficient motion monitoring data was acquired for nine fractions among nine patients.These may have been related to mistiming between the acquisition of motion monitoring and the 3D T2w scan, which was supposed to end simultaneously with the dose delivery.In total, 117 fractions of cine data were included in the study, and data from at least two fractions per patient was available for analysis.The median (range) number of cine images per fraction used to estimate the motion during treatment was 1868 (974-4610), corresponding to 6 (3-15) minutes of motion monitoring.
The population median (range) 3D root-mean-square position error was 2.9 (1.9-5.6)mm in patients treated without a compression belt and 2.1 (1.2-3.5)mm where the belt was used, which is a statistically significant difference (P < 0.01).The overall target respiration motion and mean target drift per fraction for each patient are shown in Fig. 3.The population median (range) of the mean per-patient overall target Fig. 1.A sample pre-treatment magnetic resonance scan and reference dose plan.The patient (number 25, see Fig. 3), a 79-year-old woman with a 7.5 ccm metastasis in the left adrenal gland shown in the transversal, coronal, and sagittal plane (A).The gross target volume (GTV), shown in red, moved 0.8 mm, 1.4 mm, and 6.6 mm in the left-right, anterior-posterior, and superior-inferior directions, respectively, estimated by pre-treatment 4DCT.The planning target volume, shown in light blue, was created by expanding the GTV by 4 mm in the left-right and anterior-posterior directions and 6 mm in the superior-inferior direction.The following organs at risk are shown: Stomach (brown), oesophagus (pink), aorta (purple), vena cava (dark blue), large bowel (dark green), and small bowel (light green).The target was prescribed 50 Gy in 5 fractions (B).The isodose lines shown correspond to 120% (thick yellow), 95% (medium yellow), 67% (thin yellow), and 40% (orange) of the prescribed dose.Dose-volume histograms for target volumes and organs at risk are shown (C).(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)respiration motion was 10.7 (7.5-16.9)mm for patients treated without a respiration compression belt, while it was 7.7 (4.2-11.0)mm for patients where the belt was used, which is a statistically significant difference (P < 0.01).The population median (range) of the mean perpatient overall target drift was 0.8 (0.3-4.0) mm for patients treated without a respiration compression belt, while it was 0.6 (0.2-1.7) mm for patients where the belt was used.However, this difference was not statistically significant (P = 0.37).
The mean online target peak-to-peak respiration motion over all fractions is plotted against the target peak-to-peak respiration motion determined using 4DCT in the three main directions in Fig. 4.There was a significant correlation between the online assessment and 4DCT-based target peak-to-peak respiration motion in all three directions (ρ LR = 0.55, P < 0.01; ρ AP = 0.62, P < 0.01; ρ IS = 0.76, P < 0.01).The median (range) difference in peak-to-peak respiration motion (Cine minus 4DCT) in the LR, AP, and CC direction was 0.5 (− 1.0-1.7)mm, − 0.1 (− 2.8-2.4)mm, and 0.3 (− 8.1-4.4)mm, respectively.Only the difference in the LR direction was statistically significant (P < 0.01).The overall consequences of motion during treatment delivery for the GTV are summarised in Fig. 5, showing the population median GTV DVH.The delivered dose was significantly lower than planned in regions from V84% to V128%.The median (range) GTV V95% and mean dose were significantly reduced from planned 98.6 (65.9-100) % to delivered 96.5 (64.5-99.9)% and planned 106.4 (96.9-112.9)% to delivered 105.8 (96.3-111.9)% due to motion, respectively (P < 0.01).
The per-patient change (delivered minus planned) in V95% and mean dose to the GTV due to target motion is shown as a function of the online peak-to-peak respiration motion amplitude, mean target drift, and GTV volume in Fig. 6.The change in mean dose and V95% were significantly correlated with mean drift (ρ = − 0.52, P < 0.01 and ρ = − 0.47, P = 0.01, respectively).The peak-to-peak respiration motion correlated with V95% (ρ = − 0.54, P < 0.01) but not with the change in mean dose (ρ = − 0.33, P = 0.07).The change in mean dose was significantly correlated with the GTV volume (ρ = 0.76, P < 0.01), but this was not the case with the change in V95% (ρ = 0.22, P = 0.24).No significant difference was found in the change in V95% (P = 0.59) or mean dose (P = 0.74) between patients treated with and without a respiratory compression belt.
The median (range) difference (delivered minus planned) D0.1 cc and D1cc of the summed dose to the luminal OAR closest to the GTV was 0.0 (− 0.7-2.1)Gy and 0.0 (− 0.9-2.4)Gy, respectively, and neither increased significantly due to motion during treatment (P = 0.28 and P = 0.33, respectively).The per-fraction and summed D0.1 cc and D1cc of the delivered dose to the OAR are plotted as a function of the daily planned dose in Fig. 7. D0.1 cc of the delivered dose was higher than the hard constraint of the OAR for several fractions.However, doing dose summation over all fractions, none of the hard D0.1 cc constraints was violated despite drift and respiratory motion.A similar trend was observed for D1cc.However, for one patient, the planned D1cc was deliberately chosen to violate the otherwise hard constraint to achieve a higher target coverage.As seen in the plot, D0.1 cc and D1cc were closer to the constraint for patients with targets situated in the left side than in the right side.

Discussion
This study characterised the target motion pattern in terms of drift and respiration-induced motion during online MR image-guided SBRT of patients treated for metastases in the adrenal gland.The results show that applying abdominal compression significantly reduce respiration but not drift motion.The motion patterns were used to estimate the dose delivered to the target and the luminal OAR closest to the GTV during radiotherapy.A statistically significant reduction in the target coverage due to target motion was found in the GTV DVH range above V84%.However, the change in V95% was lower than 4%, and the change in mean dose was lower than 2% for the vast majority of patients.A more considerable dose degradation was found for a few patients showing large drift or respiration motion and mainly treated without abdominal compression.Notably, the intra-fraction motion did not significantly increase D0.1 cc and D1cc to the luminal OAR closest to the GTV and did not increase the number of constraint violations when the dose was summed over all fractions.
The reduced respiration motion found in patients treated with the belt did not result in lower target dose degradation regarding the change in GTV V95% or mean dose.However, this may result from adequate PTV margins determined for the individual patient from the respiration motion measured by 4DCT.Thus, although using the belt did not result in lower target dose degradation, the reduced respiration motion will have resulted in smaller radiation fields and, thereby lower dose to the surrounding tissue.Furthermore, fewer patients using the belt experienced substantial drift or respiration motion, hence a lower excessive target dose degradation.Using abdominal compression in future patients may thus reduce the risk of a geographical miss and raise Fig. 4. Online and 4DCT respiration motion.The mean online target peak-to-peak respiration motion over all fractions plotted against the target peak-to-peak respiration motion determined in the pre-treatment phase using 4DCT in the three main directions.confidence in reaching local control.In addition, a recent study has shown that respiratory motion may change the target's apparent position in 3D MRI scans [27].Using the belt will reduce this effect for patients with targets residing a relatively long time in the exhale phase.In the current study, any image artefacts resulting from specific respiration patterns have not been taken into account.
The generation of the motion profiles was based on an in-house developed tracking algorithm that estimates the displacement of the target in interleaved coronal and sagittal cine MR images acquired during treatment delivery.The validation of the algorithm using a 4D motion phantom show promising results.However, in patients, the conspicuity of the target varies and may be more challenging to track correctly.Occasionally, the algorithm overestimated the displacements in the LR direction due to pulsatile flow-induced changes in signal from the large vessels close to the target.The peak-to-peak respiration motion in the LR direction found by the tracking algorithm was slightly higher than the motion assessed by 4DCT.This is likely also related to the pulseinduced changes from the large vessels but could also originate from the different methods used to estimate the motion.
After finalising the current research project, the tracking algorithm has been implemented in the online clinical workflow of patients treated at the MRI-linac at our department.For patients with metastases in the adrenal gland, tracking is started once recontouring of the GTV has finished and is continued throughout the plan adaptation procedure and treatment delivery.The algorithm determines the drift values to trigger an additional 3D verification scan during plan adaptation or beam interrupts during treatment delivery.
A limitation of the current study is that during the dose reconstruction it was assumed that targets and OARs moved rigidly without rotations and deformations and that patient deformations and density changes on a large scale during the treatment fraction had a negligible effect on the delivered dose.Furthermore, it was assumed that interplay effects between respiratory and multileaf collimator motion were averaged out because the treatment time was much longer than the respiration time due to the high fractional dose (7.5-15 Gy) and the relatively low dose rate of the Unity MRI-linac (~425 MU/min) [28].A similar approach has been employed previously to estimate the impact of gating on the dose delivered during MR image-guided radiotherapy of prostate cancer [29,30] as well as of liver and lung lesions [31].Menten et al. used a more advanced method to reconstruct the delivered dose to the prostate by creating multiple isocenter plans in a research version of Monaco to account for interplay effects [25].A similar approach was used by Grimbergen et al. in a study of the dose impact of intra-fraction motion during MR image-guided SBRT with abdominal compression for pancreatic and peripancreatic tumours [32].Their study showed that intra-fractional motion only modestly impacted the dose to targets and OAR.
Another limitation of the current study is that the estimated dose delivery is only an approximation since motion monitoring acquisition was not timed exactly with beam delivery, and target motion may have occurred after finishing the 2D-cine image acquisition.Furthermore, estimating the accumulated delivered dose to the target and OAR was only based on simple DVH summation and not a more advanced dose accumulation workflow based on deformable image registration [33,34].In that sense, the dose estimates are "conservative" (an upper estimate of the possible dose changes) since it was assumed that hot and cold spots were located in the same structure region at each treatment fraction.
The relatively small reduction in GTV dose coverage due to target motion is reassuring, considering the swift dose fall-off outside the GTV (PTV only covered by 67% of the prescribed dose) and no use of gating for these patients.It is currently unknown what the GTV dose reduction would have been if gating had been employed during treatment delivery.In MRI-linac systems where beam gating is currently possible, a gating window of 3 mm is typically used [18,30,31,35,36].Furthermore, a percentage of the target is typically allowed to reside outside the gating window without triggering a beam hold, and values of 5% or above are typically used [16,18,30,31,[35][36][37][38].In the study of Erhbar et al., patients treated for lesions in the lung and liver using breath-hold beam gating and a PTV margin of 5-6 mm, had a residual root-meansquare positional error of 2.4 mm inside the gating window [31].The positional target error in the current cohort of patients with targets in the adrenal gland was 2.1 mm, where the abdominal compression belt was used.Furthermore, very similar PTV margins were used in this group of patients (4-6 mm for 16 out of 17 patients).A typical patient treated with a compression belt in the current cohort might not benefit from gating when used in the typical setting, although exception gating would be valuable for cases with large drift.However, a more detailed investigation of simulated workflows is needed to estimate the fraction of patients for whom gating would be beneficial in terms of target coverage.Automatic gating on the Elekta Unity 1.5 T MRI-linac has received US FDA clearance and will likely be available for patient treatment in the coming years.

Conclusion
The target coverage reduction as a consequence of intra-fractional motion was low for most patients, and only a few patients with large drift or respiration motion experienced large target V95% or mean dose reductions.Furthermore, the motion did not lead to increased luminal OAR constraint violations.Applying abdominal compression techniques can reduce respiration motion, while controlling drift may be more challenging.Abdominal compression is recommended as motion management for adaptive radiation ablation of adrenal gland metastases at MRI-linacs without gating options.Further research is needed to investigate the benefit of gating compared to abdominal compression techniques in such patients.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 2 .Fig. 3 .
Fig. 2. Sample output data of the tracking algorithm from two patients.Left-right (LR), anterior-posterior (AP), and inferior-superior (IS) target displacement values are shown for a patient with typical drift and respiration motion amplitude on the left and a patient with large drift and respiration amplitude on the right.Raw displacement curves are shown in black, and drift is shown in red.A limited timeframe is shown to distinguish the individual peaks of the respiratory motion.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5 .
Fig. 5. Population median gross tumour volume (GTV) dose volume histogram (DVH) of the planned dose and estimated delivered dose accounting for tumour motion during treatment delivery.The shaded areas indicate the interquartile range.The p-value curve indicates regions where the curves differ significantly.

Fig. 6 .
Fig. 6.The per-patient change in V95% and mean dose to the GTV as a function of the mean online overall peak-to-peak respiration motion amplitude, the overall mean target drift, and GTV volume.Target location and use of abdominal compression indicated.

Fig. 7 .
Fig. 7.The per-fraction and summed per-patient D0.1 cc and D1cc of the delivered dose to the OAR closest to the GTV as a function of the daily adapted planned dose.In all cases, the values relative to the dose constraint are shown.Target location and use of abdominal compression indicated.

Table 1
Patient and treatment characteristics.

Table 3
Organ at risk dose constraints.The proximal kidney's constraints were soft if the distal kidney was functioning.