Intrafraction 4D‐cone beam CT acquired during volumetric arc radiotherapy delivery: kV parameter optimization and 4D motion accuracy for lung stereotactic body radiotherapy (SBRT) patients

Abstract Purpose Elekta XVI 5.0 allows for four‐dimensional cone beam computed tomography (4D CBCT) image acquisition during treatment delivery to monitor intrafraction motion. These images can have poorer image quality due to undersampling of kV projections and treatment beam MV scatter effects. We determine if a universal intrafraction preset can be used for stereotactic body radiotherapy (SBRT) lung patients and validate the accuracy of target motion characterized by XVI intrafraction 4D CBCT. Methods The most critical parameter within the intrafraction preset is the nominal AcquisitionInterval, which controls kV imaging acquisition frequency. An optimal value was determined by maximizing the kV frame number acquired up to 1000 frames, typical of pretreatment 4D CBCT. CIRS motion phantom intrafraction phase images for 16 SBRT beams were obtained. Mean target position, time‐weighted standard deviation, and amplitude for these images as well as target motion for three SBRT lung patients were compared to respective pretreatment 4D CBCTs. Evaluation of intrafraction 4D CBCT reconstruction revealed inclusion of MV only images acquired to remove MV scatter effects. A workaround to remove these images was developed. Results AcquisitionInterval of 0.1°/frame was optimal. The number of kV frames acquired was 567–1116 and showed strong linear correlation with beam monitor unit (MUs). Phantom target motion accuracy was excellent with average differences in target position, standard deviation and amplitude range of ≤0.5 mm. Target tracking for SBRT patients also showed good agreement. Evaluation of phase sorting wave forms showed that inclusion of MV only images significantly impacts intrafraction image reconstruction for patients and use of workaround is necessary. Conclusions A universal intrafraction imaging preset can be used safely for SBRT lung patients. The number of kV projections with MV delivery parameters varies; however images with fewer kV projections still provided accurate target position information. Impact of the reconstruction workaround was significant and is mandated for all 4D CBCT intrafraction imaging performed at our institution.

information. Impact of the reconstruction workaround was significant and is mandated for all 4D CBCT intrafraction imaging performed at our institution. Elekta has made 4D CBCT commercially available for clinical use through their x-ray volume imaging (XVI) system (Elekta Oncology Systems Ltd, Crawley, UK), specifically their Symmetry Vol-umeView TM module. With Symmetry, the breathing signal necessary for respiratory correlation is extracted from the CBCT projection data itself rather than from an external surrogate 4,5 or implanted fiducial markers. 6,7 There are a variety of methods for directly extracting the respiratory trace from the CBCT projections as described in detail by Yan et al., 8 however, the method used by the Elekta XVI software is known as the Amsterdam Shroud (AS) method. 1,9 This method enhances the superior-inferior motion of the internal anatomy by converting the 2D projection images into an AS image. Once generated, the breathing signal is extracted from the AS image and this signal is used to guide the retrospective binning process necessary for 4D CBCT reconstruction. Accuracy of this method relies on a distinguished moving high-density anatomical feature such as the patient diaphragm being present in the projection images. 1,8,10 Parameter settings that direct the XVI software on how to acquire the kV images as well as reconstruct them are specified by what Elekta calls acquisition and reconstruction presets. Optimal parameter settings will vary depending on the anatomical treatment site as well as the type of imaging being acquired (e.g., 3D vs 4D CBCT). Elekta provides its clinical users with a suite of default presets which have been clinically validated and optimized using patient data from their clinical partners. 11 The ability to use a single preset for multiple patients of a given treatment site and imaging modality is attractive from both a clinical workflow as well as a safety perspective. For 4D CBCT acquisitions, the default preset that is provided by Elekta is called the Symmetry preset. This preset has a significantly slower gantry speed than is used for 3D CBCT to ensure that enough projections can be acquired at the different phases of the breathing cycle to generate an accurate AS image, while minimizing artifacts in the multiple phase-based CBCT datasets. Acquisition times for 4D CBCT images are on the order of 3-4 vs 1 min for 3D CBCT acquisitions. 12 The precision requirements of lung SBRT treatment have resulted in institutions acquiring multiple CBCT datasets for each fraction: (a) 4D CBCT before treatment to measure and correct any misalignment of the time-weighted mean tumor position, (b) 3D CBCT after couch correction to verify the correct couch delta was applied and (c) 4D CBCT after treatment to assess any patient intrafraction motion. 3 Given the slower acquisition time of 4D CBCT, these additional scans can significantly increase the total treatment time for the patient. Recently, Elekta has introduced a new module to their XVI 5.0.2 software called intrafraction imaging. The XVI kV panel continues to take images at a constant frame rate of 5.5 frame/s during beam on time. Use of intrafraction CBCT allows clinicians to assess patient intrafraction stability without a post-treatment 4D CBCT and subsequently decreases the total treatment time for the patient. Since the term "intrafraction" imaging has been used in the literature for a variety of imaging techniques in assessing patient intrafraction stability we wanted to define some terms that will be used throughout this work. The term "4D CBCT" will be used to refer to all conventional 4D CBCT acquired without MV delivery.
The term "intrafraction 4D CBCT" will be used to refer to 4D CBCT acquisition with the MV beam on.
In order to acquire intrafraction 4D CBCT images, some additional preset parameters need to be defined, one of which is the acquisition interval. This parameter specifies the number of degrees the gantry must move between kV image acquisitions and essentially tells the XVI software when to acquire an imaging frame which will be used for reconstruction. This parameter combined with the gantry speed control the number of imaging frames that the XVI software can acquire per second up to the limit set by the kV detector panel. 13 A second preset parameter that will be affected during intrafraction imaging is the gantry speed setting. Previously, it was LIANG ET AL. | 11 mentioned that for the Symmetry preset, a significantly slower gantry speed was chosen to ensure that enough imaging frames were acquired at each phase of the breathing cycle to generate an accurate 4D CBCT image set. However, with intrafraction imaging, this setting must be disabled to allow the MV delivery beam settings to control the gantry speed. This poses a challenge for intrafraction 4D CBCT imaging, where the gantry speed for certain treatment deliveries may result in significant differences in the angular spacing between projections as well as an inadequate number of projections for each phase of the breathing cycle. Large and irregular angular spacing between projections has been noted to be a problem for conventional 4D CBCT acquisitions that have a uniform gantry speed. These irregularities are attributed to the sinusoidal nature of patients breathing curves and the variations in time that a patient spends in each of the 10 phases of their breathing cycle combined with the amount of gantry movement that occurs between projections at a given phase of the cycle. This results in projection clustering for some phases of the breathing cycle and missed projections for other phases. 14,15 To get around this challenge, individual intrafraction presets may need to be set for each SBRT lung patient in order to ensure an that appropriate intrafraction 4D CBCT image can be generated. This poses some clinical workflow and safety challenges, resulting in a very large database of image acquisition presets for the therapists to select from and the potential risk of choosing an incorrect preset. Another challenge to intrafraction imaging is the effects of MV scatter photons on the image quality, as the MV beam is on during kV imaging 16 . To address this, there have been a number of potential solutions which have been proposed. [17][18][19][20] Specifically, for XVI 5.0.2, Elekta has implemented a correction method where images are acquired on the kV panel during treatment delivery with the kV x-ray tube off. The MV scattered radiation noise and artifacts acquired during these MV only images are then subtracted from the kV-MV imaging frames at nearby gantry angles. By subtracting out the noise and artifacts caused by the treatment beam, Elekta states that the acquired intrafraction images are almost the same kV image acquired without the MV treatment beam on. 13,21 In this work, we use a commercial phantom to systematically validate the accuracy of target motion determined by the Elekta XVI intrafraction 4D CBCT imaging module for SBRT lung patients. To date, most of the previous reports on intrafraction CBCT have been focused on the feasibility and development of this technology and evaluation of diagnostic image quality parameters (e.g. contrast-tonoise ratio). [17][18][19][20]22,23 While there have been a few reports that have discussed the clinical implementation of this technology, 24-26 none of these studies have evaluated the accuracy of target tracking to the detail that is presented in this work. We also investigate whether a standard universal intrafraction imaging preset could be used for all SBRT lung patients by evaluating preset parameters for a wide range of arc lengths and monitor unit (MU) deliveries. Lastly, we evaluate the target motion determined from intrafraction 4D CBCT phase images for a small cohort of SBRT lung patients treated in our clinic.

2.A | Patient data
Treatment plans from 16 SBRT lung patients previously treated in 2017 at our institution were selected for this IRB-approved retrospective study. All patients were treated on an Elekta Synergy Beam Modulator machine (Elekta, Stockholm, Sweden) with 2 volumetric modulated arc therapy (VMAT) beams with arc lengths ranging from 150-200 degrees and monitor units (MUs) ranging from 842 to 2084 per beam. Since our clinical workflow would involve acquiring intrafraction 4D CBCT images during delivery of the second treatment beam, the second treatment beams from each patient plans were used to evaluate imaging preset parameters described in II.B. and to validate the intrafraction imaging module described in II.C. for lung SBRT treatments.

2.B | Development of intrafraction imaging preset for XVI
In order to use the XVI intrafraction imaging module, an acquisition preset needed to be created. Since the goal of this study was to validate intrafraction imaging for SBRT lung patients, the 4D CBCT Symmetry preset was copied and modified as per vendor recommendations. Modifications included removal of the gantry speed setting from the preset, adjusting the gantry start and stop angles for image acquisition, and the addition of three new parameters (as described below) that are specific to intrafraction imaging. 11,13,21 The first parameter IntrafractionImaging had to be set as increased the number of static kV projections acquired at the gantry start angle, 2) resulted in fewer projections acquired at the end of MV delivery, and 3) produced too many projections for large MU beams, a value of 0.1°/frame was deemed optimal for use in each of the intrafraction imaging presets that would be used in this study.
As per our past clinical gantry setting for most lung SBRT patient plans, in total four intrafraction presets as shown in Table 1 were created in our institution for SBRT to allow imaging of right and left sided lung tumors and to scan clockwise and counter-clockwise.
Those preset names must be added to machine characterization within Mosaiq system. This allows for each of the presets to be automatically selected for use within the Mosaiq/XVI Synergistiq system when a beam was selected.

2.C | CIRS phantom and validation of Elekta intrafraction imaging module
Accuracy of target motion as determined from intrafraction 4D CBCT imaging was assessed using the CIRS Dynamic Phantom (CIRS, Norfolk, VA) shown in Fig. 1. A 1 cm thick bolus was attached to the free end of the lung equivalent cylindrical rod within the phantom (as indicated by the red arrow in Fig. 1) to simulate the high density region of a patient's diaphragm. This will help the XVI software to improve the phase sorting accuracy.
The impact of this bolus addition on the intrafraction 4D CBCT phase sorting process is shown in Fig. 2. Without the bolus, discrepancies of phase sorting can be seen in the projections when the gantry angles were around −160°and near −40°to 5°. For this study, two different sized spherical targets with diameters of 10 mm (S10) and 20 mm (S20) were used and are representative of the typical tumor sizes treated for SBRT lung patients. 27,28 The target material was soft-tissue equivalent with linear attenuation within 1% of water.Target motion was programmed to follow a sinusoidal pattern as described in Eq. (1), which represents a normal patient breathing trace. In Eq. (1), Z(t) is the position of the target in the superior-inferior direction at time t, A represents the motion amplitude in mm and t represents the time in seconds. 29 Motion amplitudes of 10 mm (M10) and 20 mm (M20) were evaluated and a breathing period of 4 s was selected.
To get reference data sets for registration of 4D CBCT images, the phantom was scanned on a Philips Big Bore CT scanner using our clinical 10-phase 4DCT protocol in conjunction with the Philips An initial 4D CBCT was acquired in XVI using the Symmetry preset.
Phantom setup position was adjusted using a dual registration method where alignment was focused on the target ITV region. After couch correction, a second verification 4D CBCT was acquired and registered in XVI. The target position exhibited on this verification 4D CBCT was used as the reference for comparison with intrafraction imaging registration data eliminating any discrepancies in target positioning that may have been caused by phantom setup error.
Each beam was delivered four times to capture imaging for the four combinations of phantom setups being tested. For each CBCT session, an internal XML data file, _frames.xml was generated before xvi starting for phase sorting process. This file contains information related to each imaging frame captured by the kV imaging panel during acquisition. A snippet of the text from one of these data files is shown in Fig. 3  T A B L E 1 Gantry start and stop angles in XVI intrafraction imaging presets for lung stereotactic body radiotherapy. For all presets, the start gantry angle also is the start acquisition angle for kV imaging.
accordingly. Once this modification was completed, reconstruction was performed using the same reconstruction preset as defined in 4D Symmetry. A dual (bone followed by soft tissue) registration method was followed. Manual registration was performed when necessary. Target position on each of the 10 CBCT phases was recorded to compute the mean target position and motion amplitude/range for comparison with the verification 4D CBCT registration results. Additionally, the shape of the target motion curve constructed from 10 intrafraction phase images was compared to that corresponding curve from the verification 4D CBCT image sets.
Accuracy of target motion shape derived from the intrafraction was assessed using a time-weighted standard deviation calculation.
Assuming that the tumor position at the i-th phase image is Zi and the total acquisition time for all the kV images in the i-th phase is Ti.
The time-weighted mean position, Zave can be calculated as, and the time-weighted standard deviation can be calculated as,

2.D | Patient data validation
The intrafraction 4D CBCT imaging module was tested further for

2.E | Impact of intrafraction 4D CBCT reconstruction workaround
To determine if the inclusion of these purely MV imaging frames could cause phase sorting and reconstruction problems, we com-

</Frame>
This line shows that this imaging frame will be used for phase sorting and image reconstruction. A value of "True" means the frame will be excluded F I G . 3. Snippet of XVI internal data file that specifies what imaging frames were acquired and if they will be used for image reconstruction and phase sorting.
where P is the measured top position of the diaphragm from a given imaging frame and Pmax and Pmin represent the maximum and minimum diaphragm positions observed for that intrafraction image.
Additionally, to evaluate the impact of XVI phase sorting on diaphragm tracking for patient images, reconstructed intrafraction 4D CBCT phase image quality was compared (with and without modification of the XVI internal file).

3.B | Motion validation using CIRS phantom
The phantom target motion accuracy determined with the Elekta intrafraction imaging module for each of the 16 SBRT treatment beams is summarized in Table 2 for S10M10, S10M20, S20M10, and S20M20 respectively. The total measured amplitude uncertainty was also sub-mm with an average difference of −0.1 ± 0.5 mm, however the range was greater than 1 mm with values between −1.1 and 1.5 mm. Average differences for each of the four setups were −0.4 ± 0.4 mm for S10M10, 0.0 ± 0.5 mm for S10M20, −0.2 ± 0.2 mm for S20M10, and 0.3 ± 0.6 mm for S20M20.  Table 2 where it can be seen that the total standard deviation uncertainty for intrafraction imaging was also sub-mm with an average difference of 0.0 ± 0.2 mm and a range in values between −0.5 and 0.5 mm. Average differences for each of the phantom setups S10M10, S10M20, S20M10, S20M20 were For reference, Table 2 also shows the calculated standard deviation and amplitude results from the reference 4D CBCTs from each of the four phantom setups as well as the nominal values calculated from Eq. (1). It should be noted that intrafraction 4D CBCT phantom measurements were taken over two separate days resulting in the two reference 4D CBCT data sets with values shown in Table 2.
Comparatively, the reference 4D CBCT data sets show good agreement with the nominal calculated values with a max difference of 0.6 mm seen for the nominal amplitude with the S10M20 phantom setup.
Plots of the measured uncertainty in phantom target tracking

Beam information
Mean position difference (mm)

S20M10
Amplitude Difference (mm) F I G . 6. Relationship between target motion uncertainty in mean position, standard deviation and amplitude as a function of MV treatment beam gantry span, and MV treatment beam MUs for each of the four phantom setups (S10M10, S10M20, S20M10, S20M20). The Y-axis title = chart title. Table 3 presents the target motion results for a single fraction for each of the three SBRT patients evaluated in this study. The difference in tumor mean position between the pretreatment 4D CBCT and the intrafraction 4D CBCT represents target motion during treatment delivery. Target tracking for each of the three SBRT patients is presented in Fig. 8 and demonstrates the similarity in the target motion shape with respect to amplitude and standard deviation observed for the pretreatment and the intrafraction 4D CBCTs.

3.C | Patient Data Validation
From Table 3, the largest difference in amplitude between the pretreatment and the intrafraction 4D CBCTs was 2.3 mm for patient B, and standard deviation differences were <1 mm for all three patients.   Fig. 10(b). Figure 11 shows the image quality of the 4D CBCT phase images for the above patient case compared with and without workaround.

3.D |
F I G . 7. Intrafraction four-dimensional cone beam computed tomography reconstructed S10M20 phantom images for three breathing phases (Top-row: phase 0, Mid-row: phase 9, Bottom-row: phase 3). Phantom target is a sphere and should appear as a circle on reconstructed phase images. Deformation of target shape as a result of unevenly spaced and reduced amount of kV projections can be seen. The Window and Level settings used in the Pinnacle Planning system were: 259 and 238.
T A B L E 3 Patient intrafraction target motion results for each of the three stereotactic body radiotherapy (SBRT) patients evaluated in this study. The ability to use a universal preset for all patients of a given treatment site greatly simplifies clinical workflow and reduces potential errors that could arise from having to manually create, associate and select a different preset for each patient. This also ensures that the images generated with that preset will be optimal for guiding patient setup for that specific treatment site, reducing the need to take subsequent CBCT images.

Patient
When implementing intrafraction imaging, the question arises as to whether a universal preset can be used since the gantry speed for intrafraction 4D CBCT acquisition depends on the gantry speed required for MV delivery. Since MV delivery gantry speeds vary depending on the complexity of the SBRT lung plan, it is quite possible that use of a standard intrafraction imaging preset could result in suboptimal 4D CBCT images for some patients.
In this study, we evaluated different settings for the Acquisi-tionInterval parameter. Elekta recommends not setting this parameter for 4D CBCT acquisitions to prevent the clinical user from using a value that is too large and generating images with not enough imaging frames. However, we found that leaving this value blank   In their study, they report a mean difference in target position of <0.4 mm and an average standard deviation of 0.6 mm. 23 Similarly, when intrafraction images were acquired for three SBRT lung patients, measured amplitude and standard deviation results (Table 3) and target tracking shapes (Fig. 8) were also found to be in good agreement with values characterized on pretreatment 4D CBCT.  In our study, the maximum kV only gantry span for the beams tested was 50°out of a total gantry span of 200°of which 58 kV projections were acquired. From Table 2, image quality for that beam was still adequate enough to determine target motion for all the phantom setups tested with a maximum absolute uncertainty of only 0.8 mm for target mean position, standard deviation and amplitude. However, intrafraction 4D CBCT images acquired for smaller MV arc lengths could be significantly impacted. This is because a large portion of the total imaging gantry span will be acquired kV-only, using the 180°/min gantry speed. This point is supported by the actual acquisition interval obtained for each of the 16 treatment beams tested. The acquisition interval for the largest MU beam was 0.18°/frame compared to 0.35°/frame for the smallest MU beam. This is a difference of about a factor of 2, meaning that for large MU beams almost double the number of kV projections will be acquired.
The varying number of kV projections for intrafraction imaging as well as the nonuniform gantry spacing at which projections are acquired was shown to affect the reconstructed phantom target shape (Fig. 7). While these changes in reconstructed tumor shape did not impact the accuracy of any of the target motion parameters assessed in this study (mean position, amplitude, and standard deviation), we expect that these changes would have an impact on any deformable image registration and deformable dose accumulation evaluations performed with these images. Further study on these effects is currently under our investigation.
Ideally, all MV only frames should be excluded from the phase sorting and reconstruction process. However, upon review of these XML files, it was found that those imaging frames were not being excluded and a reconstruction workaround was developed. The impact of this workaround was assessed for phantom and patient images. Errors in phase sorting for phantom images without the workaround were quite limited. The insignificance of including the MV frames in the phase sorting process for phantom images may be attributed to the construction of the lung phantom which has significantly higher contrast differences between the target volume and high-density bolus (pseudo-diaphragm) materials against the adjacent lung phantom material than is seen in lung patients. This allows for a reasonably accurate phase sorting process and reconstructed intrafraction 4D CBCT. However, errors in phase sorting for patient images without the workaround were significant [Figs. 10(a) and 10(b)] resulting in blurring of the diaphragm position in the reconstructed 4D CBCT phase images (Fig. 11), all of which were corrected through the implementation of the workaround. This example strongly demonstrates that modifying the XVI internal file, _frames.xml prior to sorting and reconstruction is necessary. Clinically, we have mandated that this workaround be performed for all intrafraction 4D CBCT imaging. The workaround is easy to handle for the therapists, and it would cost an additional one minute patient treatment time.

| CONCLUSIONS
We have conducted a systematic study of the Elekta XVI intrafraction 4D CBCT imaging module for SBRT lung patients and validated this imaging technique with phantom and patient measurements.
Through this work, we determined that a standard intrafraction imaging preset using an AcquisitionInterval parameter of 0.1°/frame can be used safely for all SBRT lung patients, greatly simplifying clin-