Evaluation of radixact motion synchrony for 3D respiratory motion: Modeling accuracy and dosimetric fidelity

Abstract The Radixact® linear accelerator contains the motion Synchrony system, which tracks and compensates for intrafraction patient motion. For respiratory motion, the system models the motion of the target and synchronizes the delivery of radiation with this motion using the jaws and multi‐leaf collimators (MLCs). It was the purpose of this work to determine the ability of the Synchrony system to track and compensate for different phantom motions using a delivery quality assurance (DQA) workflow. Thirteen helical plans were created on static datasets from liver, lung, and pancreas subjects. Dose distributions were measured using a Delta4® Phantom+ mounted on a Hexamotion® stage for the following three case scenarios for each plan: (a) no phantom motion and no Synchrony (M0S0), (b) phantom motion and no Synchrony (M1S0), and (c) phantom motion with Synchrony (M1S1). The LEDs were placed on the Phantom+ for the 13 patient cases and were placed on a separate one‐dimensional surrogate stage for additional studies to investigate the effect of separate target and surrogate motion. The root‐mean‐square (RMS) error between the Synchrony‐modeled positions and the programmed phantom positions was <1.5 mm for all Synchrony deliveries with the LEDs on the Phantom+. The tracking errors increased slightly when the LEDs were placed on the surrogate stage but were similar to tracking errors observed for other motion tracking systems such as CyberKnife Synchrony. One‐dimensional profiles indicate the effects of motion interplay and dose blurring present in several of the M1S0 plans that are not present in the M1S1 plans. All 13 of the M1S1 measured doses had gamma pass rates (3%/2 mm/10%T) compared to the planned dose > 90%. Only two of the M1S0 measured doses had gamma pass rates > 90%. Motion Synchrony offers a potential alternative to the current, ITV‐based motion management strategy for helical tomotherapy deliveries.


| INTRODUCTION
Intrafraction motion limits the conformity of radiation therapy treatments by prohibiting tight margins around the clinical target volume (CTV). Without compensating for motion during treatment, sufficiently large planning target volumes (PTV) are necessary to minimize the risk of underdosing the target while also resulting in a larger volume of normal tissue irradiated. This expanded volume is commonly referred to as the internal target volume (ITV). 1 In addition, interplay between the motion of the target and motion of the collimation can cause undesired dose distributions inside the PTV, especially for hypo-fractionated treatments and cannot be accounted for with a margin. 2 Respiratory motion is a primary source of intrafraction motion for treatment sites in the thorax and abdomen. 3,4 Motion management is recommended by the AAPM Task Group 76 (TG-76) for respiratory motion >5 mm in any direction. 5 Non-ITV motion management techniques include gating and tracking, which both require precise knowledge of tumor location during treatment. 6 Unfortunately, the precise location of the tumor during treatment is difficult to obtain. Target localization relies on internal or external surrogates, which can be determined based on optimal surface monitoring, radiofrequency beacons, kilovoltage (kV) x-ray imaging, or magnetic resonance imaging (MRI). 3,7,8 CyberKnife ® (CK) uses the Synchrony ® Respiratory Tracking system (Accuray Incorporated, Sunnyvale, CA), which combines external surrogate monitoring and x-ray imaging of implanted fiducials to model and predict tumor motion due to respiration. [9][10][11] In this system, the robotic movements of the CK delivery system are adapted in real time to compensate for motion.
Radixact ® (the next-generation TomoTherapy ® System; Accuray Incorporated, Sunnyvale, CA) is a helical tomotherapy radiation therapy delivery system capable of delivering conformal intensitymodulated radiation therapy (IMRT). 12 The continuous couch and gantry motion during treatment complicates conventional gating techniques. The Radixact contains an intrafraction motion management system called Synchrony ® , which has been adapted from CK Synchrony. 13 On the Radixact, an x-ray tube and flat-panel kV imager are offset 90°from the megavoltage (MV) imager and beam, shown in Fig. 1. The kV imaging subsystem is used to periodically localize the target during treatment (while the gantry is rotating). Two kV radiographs are separated in time to allow the gantry to rotate. Therefore, sequential monoscopic images provide delayed stereoscopic information. For monitoring respiratory motion, light-emitting diodes (LEDs) are placed on the patients' chest and identified with a camera mounted to the ceiling to provide the phase of respiration, shown in Fig. 1. The target can be localized with or without implanted fiducials near the target, but this work will only consider the fiducial-based respiratory Synchrony option. Schnarr et al. described the modeling of the target location based on the information from the kV radiographs and the external LEDs. 13 The respiratory model is used to change the existing jaw locations and multi-leaf collimator (MLC) leaf openings in real time during treatment. The model is updated every time new kV radiographs are acquired, without interrupting the treatment. Therefore, treatment time is the same for a Synchrony treatment as a conventional treatment unless the delivery has to be paused to acquire additional images to improve the motion model. The jaws compensate for target motion in the IEC-Y (superior/inferior when head-first supine) direction. Target motion in the IEC-X (left/right) and IEC-Z (anterior/ posterior) directions is compensated by changes in the leaf opening positions. For example, when the gantry is at 0°(central axis along the IEC-Z direction), motion in the IEC-X direction is compensated by changing the leaf openings; motion in the IEC-Z direction is not compensated since the target is moving along the line of the beam.
Likewise, when the gantry is at 90°, the MLC leaf openings compensate for target motion in the IEC-Z direction.
The dosimetric effect of a Synchrony-enabled treatment for respiratory motion was studied by Schnarr et al., in which Synchrony tracked and compensated for linear respiratory-mimicking motion, but for only one non-clinical treatment plan for a cylindrical target. 13 In addition, Chao et al. evaluated the accuracy of TomoTherapy dose calculations for Synchrony deliveries using film, but the motion was programmed a priori into the treatment plan instead of using realtime tracking. 14 16 The effects of separate target and surrogate motion on tracking accuracies and dosimetric fidelity of Radixact Synchrony have not been investigated.
The current work investigates the Synchrony system on Radixact using real-time tracking and modeling for realistic three-dimensional (3D) respiratory motion and clinical IMRT plans. Two aims will be addressed: (a) evaluate the ability of the Synchrony system to accurately model and track motion of a phantom moving according to simulated respiratory motion, and (b) perform patient-specific DQA to assess the deliverability of Synchrony plans correcting for undesired effects of intrafraction respiratory motion on helical tomotherapy treatments.

| MATERIALS AND METHODS
Imaging datasets and treatment planning orders were selected for subjects with abdominal or thoracic tumors enrolled in an IRB-approved study. Subject cases with lung, liver, and pancreas targets were selected to be replanned for delivery on a research Radixact system for this motion management study. Table 1 shows treatment information of each of the 13 subjects. All subjects had 4DCT scans as part of their clinical simulation, where they were instructed to breathe normally. The CTV was delineated on the maximum inspiratory breath hold (MIBH) image. FERRIS ET AL.

2.A. | Motion traces
Motion traces were generated uniquely for each subject in MATLAB ® (MathWorks, Inc., Natick, MA). Characteristics of the motion traces used in this work for the 13 subjects are shown in Table 2. Motion was modeled using an equation proposed by Lujan et al., where A is the amplitude, t is time, T is the period, φ is the starting phase in radians, and n is a fitting parameter to model more time spent at exhale than inhale. 17 Equation (1) was used to separately model the X, Y, and Z motion of the target and the Z motion of the patient's chest, or the surrogate (S). The period and amplitude for each respiration were randomly sampled from a normal distribution.
The mean and standard deviations of these normal distributions were chosen using values that are typical of patient respirations observed in the literature, such as a mean period between 3 and 5 s. 3,4 The period and amplitude of subsequent respirations were smoothed using a moving average filter to prohibit sudden changes in these parameters between respirations. Hysteresis in the sagittal plane was modeled by a phase shift in the Z direction (φ Z ) relative to the Y direction, which was constant throughout a given subject's motion. Phase shifts between target motion and surrogate motion were modeled by φ S , which was also constant throughout a given subject's motion. A positive phase shift for the S or Z motion indicates that the motion in that direction is delayed. Shifts are specified in terms of percent: a shift of /2 is denoted a 50% phase shift. The fitting parameter (n) was chosen to range from 1 to 3 in this work based on the observations of Seppenwoolde et al. 18 . A shift in baseline positionmodeled by a low-frequency, low-amplitude cosine was incorporated into the traces since these shifts have been observed to have the greatest effect on helical tomotherapy treatments. 19 The baseline shifts throughout treatment ranged from 0 to 2.5 mm.

2.B. | Treatment planning
For each subject, a helical tomotherapy plan was generated in the   to-peak and the target is aligned to the mean position. There is no mechanical limit on amplitude of motion in the IEC-X and IEC-Z directions (other than extreme off-axis targets), as motion in these directions are compensated by MLC leaf openings. The 2.5 cm jaw setting was used for all plans in this study.

2.C. | Validation measurements
Delivery validation of these clinical patient plans was performed using a customized Phantom+ and a Hexamotion ® stage (ScandiDos Inc., Uppsala, Sweden), shown in Fig. 2. The Hexamotion stage provided 3D translational movements described by the generated motion traces. The Phantom+ was modified to house a CyberKnife "ball-cube" insert with embedded fiducials (Fig. 2). These fiducials are imaged with the kV radiographs and are used as a surrogate of target position.
Two LED placement locations were used in this work. For the 13 subject cases, the LEDs were placed on the Phantom+ itself. For one subject case (Lung 5), additional investigations were performed with the LEDs placed on a separate, 1D surrogate stage in front of the Phantom+, shown in Fig. 2. This setup was used to explore the effect of varying the relationship between the surrogate LED motion and the internal target motion, as described by Akino et al. 16 Characteristics of the motion traces generated for each subject case. The parameters are in reference to Eq. (1). The fitting parameter (n) was 2 for all cases in this table. The X and Y directions were always in phase (φ X = 0). The RMS displacement from the origin is a metric used to describe the 3D magnitude of motion from the phantom origin location (where it was registered). With motion tracking turned off, δ RMS for each case in Table 3 would be equal to this value. unclear. Therefore, the M1S0 case was used to benchmark the impact that the motion had upon the static plan. indicate that for 95% of the treatment time, the error between phantom motion and the tracked motion was 1 mm or less. The median dose difference was intended to quantify dosimetric differences inside the target region, therefore a threshold was applied considering points above 50% of the maximum dose. All dose comparison metrics were calculated using the Delta4 software. One-dimensional profiles in the X, Y, and Z directions were extracted from the Delta4 software to analyze the shape of the dose distributions. Table 3 shows tracking error statistics between the Synchrony-predicted motion and the phantom motion for each of the 13 M1S1

3.A. | Motion Tracking
subject cases with the LEDs on the Phantom+. The ratio of respiratory period to the average time between the kV radiographs, or "images per respiration," varied from 0.4 to 1.6, as shown in Table 3.
The values of δ RMS were 1.5 mm or less for all cases and the values of δ 95% were <3.0 mm for all cases. Motion traces and tracking error plots for three example subject cases are shown in Fig. 3. Table 4 shows tracking error statistics between the Synchronypredicted motion and the phantom motion for each of the cases of Lung 5 with the LEDs on the Phantom+ or on the surrogate stage.
The original case with the LEDs on the Phantom+ is termed "Lung 5" and the additional cases with the LEDs on the surrogate stage were termed "Lung 5a"-"Lung5i." For Lung 5b-Lung 5i, only one parameter was changed relative to Lung 5a at a time. The motion parameters of Lung 5a and the changes of the subsequent cases are shown in Table 4. Figure 4 shows target and surrogate motion traces for Lung

3.B. | Dosimetry
Dosimetric analysis between each of the three measured doses and the planned dose for the 13 cases with LEDs on the Phantom+ is shown in Table 5. All the M0S0 and M1S1 measured dose distributions and only two of the M1S0 measured dose distributions had gamma pass rates (3%/2 mm/10%T) above the universal action limit (90%) as recommended by TG-218. 24 All the M0S0 and M1S1 measured dose distributions and eight of the M1S0 measured dose distributions had a median dose difference (50% threshold) compared to the planned dose within AE3%.

4.A | Motion tracking
If the ratio of the time between images to the respiratory period is one, that means that exactly one image is acquired per respiration.
As a result of this, aliasing can occur between the kV images and the respirations, that is, the images may be acquired only at one phase of the breathing cycle. This can lead to uncertainties in target position during the other phases of the breathing cycle. For Liver 1, this was initially observed when using four imaging angles per gantry rotation, shown in Fig. 6. The average breathing period was 3 s and the gantry rotation period was 11.8 s, therefore the number of images per respiration was approximately 1.0. The model was not T A B L E 4 Dosimetric and tracking analysis for the various LED motions for the Lung 5 case with separate LED and phantom motion. The parameters of the Lung 5a case were used as the base parameters: mean surrogate amplitude (ÀA S ) of 5 mm, phase shift (φ S ) of 0%, fitting parameter (n) of 2, and mean period (ÀT) of 5 s. For Lung 5b-5i, only one parameter was changed relative to Lung 5a. Gamma pass rates are for the M1S1 case. The pass rates for the M1S0 case for Lung 5 are in Tables 5 and 6.

Case LED location
Parameter change from Lung 5a γ pass: 3%/2 mm/10%T (to plan) The same fitting parameter (n) was used in Eq. (1) for both the LED and target motion when the fitting parameter was changed. The amplitude and period of motion of both the LED and target motion were unchanged when this parameter was changed.
F I G . 4. (a) Full-treatment and 30-second sample phantom motion trace for Lung 5a in the X, Y, Z, and surrogate directions. The X, Y, Z motion of the target was not changed for cases Lung 5b-Lung 5i (other than fitting parameter change for Lung 5h and Lung 5i). (b) Cumulative tracking error plots showing the probability of observing a tracking error greater than the specified value throughout the treatment for various cases.
able to be accurately built with these parameters. The plan was changed to have five images per gantry rotation (~1.3 images per respiration), and the M1S1 plan was successfully delivered. Figure 6 shows that the kV images sampled all phases of respiration more evenly with five images per gantry rotation than four. The number and angles of kV images per gantry rotation can be chosen prior to the treatment and can be modified during the treatment if aliasing is observed. Breath-coaching could be used to avoid respiratory frequencies aliasing with the imaging frequencies.
In this work, the number of images per respiration was not found to be correlated with tracking accuracy as long as this value was not near unity. For example, this value was the smallest at 0.4 for Liver 4, and the RMS error between the Synchrony trace and the phantom trace was 0.8 mm. Images were acquired only once every 2.5 respirations, but this work indicates that as long as respiration is regular during this time, the model will stay relatively constant and does not require more frequent updating. Future work will investigate the limits of imaging frequency and accuracy of the resulting model. dose were all greater than 98% (Table 6), suggesting motion management would not have been needed in these cases for a gammabased DQA.
The profiles for Lung 2 were included in Fig. 5 to show that even though gamma pass rates were high without motion management, dose blurring can be observed in the M1S0 profiles in the superior/ inferior and anterior/posterior directions which is not observed in the M1S1 profiles. It is unclear whether these changes in the profiles necessitate motion management, but it is clear that when motion management was used, the resulting profiles more closely resembled the static profiles.
Population statistics were intended to indicate whether phantom motion has a significant impact on the dose distribution when compared to the M0S0 dose and whether motion synchronization corrects for this. Although this is a small subset of cases (n = 13), there was evidence suggesting a difference between using Synchrony (M1S1) and not using Synchrony (M1S0) on the dosimetric similarity to the static measured dose (M0S0) for the metrics shown in Table 6. Motion Synchrony did not have a large benefit for every subject in this work, but these statistics indicated that for these subjects and metrics, there was an overall significant improvement when Synchrony was used compared to when it was not used.

| CONCLUSION
The motion Synchrony system on the Radixact was able to track and synchronize the delivery of radiation with realistic 3D respiratory motion of a phantom for clinical helical tomotherapy treatments.
When motion Synchrony was used with a moving phantom, the measured dose distribution more closely matched the planned dose.
The impact of motion on the treatments was found to vary from subject to subject, but overall, there was evidence suggesting significant improvement in agreement with the static dose distribution when using Synchrony. In addition, motion interplay and dose blurring effects were not observed with Synchrony enabled. The sensitivities of CK Synchrony to surrogate phase shifts reported in the literature 16 were not observed for the Radixact Synchrony system.
This work provided evidence that Synchrony reduces effects of intrafraction respiratory motion that are not accounted for using the current, ITV-based motion management strategy for helical tomotherapy plans.
F I G . 6. 30-s sample of Synchrony-modeled motion ("Tracking") vs phantom motion in the IEC-Y direction for Liver 1 with four images per gantry rotation (top) and five images per gantry rotation (bottom). Multicolored points indicate the phase at times which the kV images were acquired (four colors in the top and five colors in the bottom). When imaging with four images per gantry rotation, the imaging frequency was found to alias with the breathing frequency and the model was not accurately built (hence there is no tracking curve in the top image).