Seven patients with a single pelvic lymph node oligometastasis previously treated using SBRT on an Elekta Unity (Elekta AB, Stockholm Sweden) were enrolled in this study. This research was reviewed by our departmental institutional review board (IRB) and all patients enrolled provided autonomous informed consent which included consent to publish. The study was conducted in accordance with the International Council for Harmonization ICH E6(R2) Good Clinical Practice as adopted by the United States FDA, which aligns with the principles of Helsinki. All patients enrolled in this study had a GTV which was well visualized on a balanced fast field echo (TE 3.8 ms, TR 1.92 ms, flip angle: 40 degrees) cine MRI imaging sequence which comes standard on the Unity system (32).

The patients treated on this study received 35 Gy in 5 fractions to the planning target volume (PTV). Reference planning was conducted on a CT dataset with 2 mm slice thickness in a research version of the Monaco treatment planning system (version 6.01). The lymph node metastasis visualized on the CT image was used for GTV delineation. No additional margin was added to create a CTV (GTV = CTV). For this study, three separate reference plans were made for each patient and fraction consisting of a different PTV margin expansion. The GTV was expanded uniformly by 2mm, 3mm, and 5mm respectively to create the PTV in each plan. A nine-field step-and-shoot IMRT beam arrangement which avoided the cryostat pipe was used for planning (9, 32). Each PTV margin plan was created using a maximum of 10 segment shape optimization loops and IMRT parameters were held constant for each plan. All plans were normalized such that 95% of their respective PTV received 35 Gy. A 1% Monte Carlo per plan statistical uncertainty was used for each optimization.

Online adaptive planning was carried out using the adapt-to shape (ATS) methodology (11, 33). A daily T1 3D image dataset was acquired on the Unity system and contours from the reference CT plan were deformed onto the daily MR image dataset. The OARs and GTV were manually edited as needed. In the clinical setting, the Hyperion optimizer in the Monaco TPS was used to generate a fully adapted plan starting from fluence (34). Following the completion of the adapted plan, online quality assurance procedures were performed and MRI cine motion monitoring images were acquired throughout the duration of treatment (32).

For this study, a simulated ATS workflow was carried out in the research Monaco software. This was done for all 35 fractions (5 fractions per patient) and for each PTV margin such that 105 treatment plans were generated in total. Each plan was normalized for 95% PTV coverage and used the same IMRT parameters as the reference plan. The minimum dose received by the hottest 0.5cc of the bowel (Bowel D_0.5cc) was compared on each daily adaptive plan as a function of the PTV margin. Violations in all other OAR objectives were assessed using dosimetric criteria previously described for pelvic oligometastases using this dose and fractionation scheme (35).

All the reported OAR doses in this study are from adaptive plans and do not consider effects of intra-fraction motion. The tracking algorithm used in this study only tracked the GTV. Other OAR structures may undergo deformations and/or not move with the same rigid motion as the GTV and thus could not be assessed.

The GTV was retrospectively tracked on the cine MR images that were acquired during each treatment fraction using a research tracking algorithm provided by Elekta. Details of the algorithm have been previously described (36). Briefly, the tracking process begins with a training phase where 30 sagittal and 30 coronal images are obtained. As this study focuses on tracking targets which do not experience respiratory motion, a single average image from the sagittal and coronal training set is generated. This sagittal and coronal image are referred to as template images. A sagittal and coronal plane through the centroid of the target tracking structure is extracted from the daily 3D MR image dataset that was used for planning. A mutual information algorithm is used to register the template images to the extracted planes of the daily 3D MR image. This component of the registration can be manually edited by the user prior to the initiation of treatment. This is referred to as the absolute registration component. Finally, live incoming cine MRI images are registered to the template image using a cross correlation algorithm which focuses registration on the tracking structure plus an additional margin expansion referred to as the binary mask. The registration of the incoming cine images to the template images is referred to as the relative registration. The total displacement of the GTV is then equal to the sum of the absolute and relative registration components. An overview of the tracking algorithm is provided in Figure 1 . This algorithm only considers rigid 3D translations and does not account for target rotations or deformations.

To help ensure accurate tracking, the algorithm also employs a quality factor metric. Some of the parameters evaluated by the metric include interplane jitter, detection of large deformations, and large through-plane motion which can alter the apparent size of the target in a given frame. In clinical practice, any frames in which the quality metric reports a failure would result in the radiation beam being turned off. For this reason, any coronal and sagittal paired frames which reported a quality factor failure were excluded from analysis. To assess potential systematic trends in the GTV position with respect to treatment duration, the average position and standard deviation of the GTV over all 35 factions was plotted at each time point recorded.

The delivered fractional dose received by the GTV was reconstructed based on the position of the GTV as determined by the tracking algorithm and the IMRT delivery. To do this, the IMRT delivery time was estimated for each treatment beam such that it could subsequently be synchronized with the MRI cine images. The treatment time per beam was estimated using a kinematic model incorporating beam delivery time and mechanical motion parameters. The Monaco TPS reports the total time required to deliver the monitor units assigned to each beam and reports the total time required to move the MLCs between segments in each beam. To obtain the total delivery time per beam, additional factors were added including the gantry rotation time between each beam angle and a correction factor which accounts for beam on delay and dose rate ramp-up per IMRT segment. Beam on delay refers to the time from which the MLCs reach their intended position until the time that the radiation beam is initiated. A correction factor for beam on delay and dose rate ramp up was determined by comparing Monaco TPS delivery estimates with actual deliveries times for pelvic oligometastatic cases. A correction factor was fitted to minimize the difference between estimated and actual plan delivery times. A value of 1.3 seconds per segment was used in this study.

Dosimetric changes to the GTV coverage were then evaluated for each patient, fraction, and margin using a custom in-house program written in MATLAB (Mathworks Inc, Natick, Massachusetts, Version 2020b). Rigid motion of the GTV was recorded for each cine frame creating a timeline of the GTV motion throughout the course of treatment that was synced with the radiation delivery. The displacement of the GTV from its planned location was taken as the average position during each beam delivery. Radiation dose was exported in a DICOM format. Instead of translating the target within a fixed beam arrangement, the isocenter location listed for each DICOM file was modified to represent the effective motion of the beam from the fixed reference frame of the GTV. Dose distributions were recalculated using a cubic spline interpolation to resample the translated dose distribution from each field back onto the original DICOM coordinate system, which was then compiled into a single radiation dose DICOM file among all modified treatment beam dose distributions. The resulting coverage to the GTV was calculated using Velocity (Varian Medical Systems Inc., Palo Alter, CA, version 3.2.1). A summary of the workflow used to determine the intra-fractional accumulated dose to the GTV is provided in Figure 2 .

Future clinical workflows will enable the ability to perform additional intra-fraction plan adaptions via a modified adapt-to-position workflow (11, 37). A three second moving average of the tracked GTV position was used to determine how many intra-fraction adaptive interventions would be needed to keep the GTV entirely within the respective PTV margin for each fraction. With this strategy, if any of the three principal directions had a tracked value greater than the PTV margin than an intervention would be required. The time point at which a baseline shift was needed would reset the target position in all three directions back to zero and the relative displacement of the target would be tracked from that point forward to determine if additional baseline shifts corrections would be needed. The number of required baseline shifts was recorded for each PTV margin plan and fraction.

All dosimetric constraints were achieved in 85.7%, 94.3%, and 100.0% of the daily ATS plans which used 5 mm, 3 mm, and 2 mm PTV margins, respectively. All the recorded violations occurred for the Bowel D_.05cc < 32 Gy constraint. OAR objectives for the bladder, rectum, and sigmoid were met for all fractions and PTV margins used based on the dosimetric criteria published by Winkel et al. (35) Compared to plans with a 5 mm PTV margin, there was a 27.4 ± 12.3% (4.0 ± 2.2 Gy) and a 18.5 ± 7.3% (2.7 ± 1.4 Gy) reduction in the bowel D_0.5cc for 2 mm and 3 mm PTV margins, respectively.

The percentage of the GTV which was planned to receive at least 35 Gy (GTV V_{35 Gy}) on the daily ATS plans was 100.0 ± 0.1%, 100.0 ± 0.1%, and 99.9 ± 0.3%, when averaged among all fractions for the 5 mm, 3 mm, and 2 mm PTV margin plans, respectively. When accounting for the effects of intra-fraction motion, the actual delivered GTV V_{35 Gy} doses were on average 100.0 ± 0.1%, 99.6 ± 1.0%, and 99.0 ± 1.4% for the 5 mm, 3 mm, and 2 mm PTV margin expansion plans. The minimum coverage in any delivered fraction was 99.6% for 5 mm PTV margins, 97.2% for 3 mm PTV margins, and 95.0% using 2 mm PTV margins. The application of baseline shift corrections was not considered in the intra-fraction dose analysis. The box and whisker plot shown in Figure 3 summarizes the GTV coverage and bowel doses received by each patient and margin combination.

The quality factor metric indicated successful GTV tracking for greater than 99.8% of the cine imaging frames in each fraction. The average and standard GTV displacement among all patients and fractions is shown in Figure 4 where each time point represents the average position among all patients and fractions at that same time point during treatment. The average GTV position remained within 1 mm of the planned position in each principal direction for all time points. Thus, the intra-fraction motion for these pelvic oligometastases appeared to be largely random with only a small systematic drift component with respect to treatment duration. While Figure 4 plots the position of the GTV during treatment, the positional drift of the GTV during adaptive planning was captured using the absolute registration value of the tracking algorithm. This is the reason why the GTV position is not zero at the onset of treatment in Figure 4 . One of the three principal motion components in the absolute registration exceeded the uniform PTV margin in 17%, 6% and 0% of the delivered fractions for the 2 mm, 3 mm, and 5 mm uniform PTV margins, respectively. An absolute registration value which exceeds the PTV margin indicates that part of the GTV volume extends outside of the PTV prior to the start of treatment. The mean and standard deviation absolute registration values for each patient averaged over all 5 treatment fractions are shown in Table 1 .

Evaluating the GTV position at the beginning and again at the end of treatment would not have adequately captured the maximum range of GTV motion for all cases. Figure 5 provides one example of this and displays the motion trace of the second fraction for patient 7. The GTV was within 2mm at the start and end of treatment in each of the 3 principal directions, but the target had an excursion of up to 4.7mm during the treatment. The target drifted outside of 2 mm in the lateral direction at approximately 68 seconds and physical patient motion happened causing shifts in the A/P and S/I directions between approximately 440 and 460 seconds. However, the patient returned to a near baseline position by 500 seconds.

While the average GTV coverage over all 35 fractions was 99.0% or greater, even when using 2 mm PTV margins, there are still instances where a portion of the total volume of the GTV can drift outside of the PTV during treatment causing reduced coverage for individual fractions. Figure 5 provides one example of this, and the data presented in Figure 3A shows that individual fractions can have GTV coverage as low as 95% in this study. The ability to perform intra-fraction baseline shift plans can improve the target coverage in these instances.

Over all fractions the centroid of the GTV never exceeded 5mm in any of the three principal directions and as such no baseline shifts would have been required when using a 5mm PTV margin. An average of 0.8 ± 1.0 and 0.3 ± 0.6 baseline shift corrections per fraction would have been needed when using 2 mm and 3 mm PTV margins, respectively. A maximum of 5 baseline shifts would have been required in a single fraction for the 2 mm margin and up to 3 corrections when using a 3 mm PTV margin. 91.4% and 97.1% of the fractions would have required 1 or fewer baseline shift corrections for 2 mm and 3 mm PTV margin expansions. Only one of the seven patients would have had any fractions requiring more than one baseline shift correction regardless of the margin used. The average number of baseline shifts required for each patient and PTV margin is shown in Table 1 .