Impact of use of optical surface imaging on initial patient setup for stereotactic body radiotherapy treatments

Abstract Purpose To evaluate the effectiveness of surface image guidance (SG) for pre‐imaging setup of stereotactic body radiotherapy (SBRT) patients, and to investigate the impact of SG reference surface selection on this process. Methods and materials 284 SBRT fractions (SG‐SBRT = 113, non‐SG‐SBRT = 171) were retrospectively evaluated. Differences between initial (pre‐imaging) and treatment couch positions were extracted from the record‐and‐verify system and compared for the two groups. Rotational setup discrepancies were also computed. The utility of orthogonal kVs in reducing CBCT shifts in the SG‐SBRT/non‐SG‐SBRT groups was also calculated. Additionally, the number of CBCTs acquired for setup was recorded and the average for each cohort was compared. These data served to evaluate the effectiveness of surface imaging in pre‐imaging patient positioning and its potential impact on the necessity of including orthogonal kVs for setup. Since reference surface selection can affect SG setup, daily surface reproducibility was estimated by comparing camera‐acquired surface references (VRT surface) at each fraction to the external surface of the planning CT (DICOM surface) and to the VRT surface from the previous fraction. Results The reduction in all initial‐to‐treatment translation/rotation differences when using SG‐SBRT was statistically significant (Rank‐Sum test, α = 0.05). Orthogonal kV imaging kept CBCT shifts below reimaging thresholds in 19%/51% of fractions for SG‐SBRT/non‐SG‐SBRT cohorts. Differences in average number of CBCTs acquired were not statistically significant. The reference surface study found no statistically significant differences between the use of DICOM or VRT surfaces. Conclusions SG‐SBRT improved pre‐imaging treatment setup compared to in‐room laser localization alone. It decreased the necessity of orthogonal kV imaging prior to CBCT but did not affect the average number of CBCTs acquired for setup. The selection of reference surface did not have a significant impact on initial patient positioning.


| INTRODUCTION
Optical surface imaging is an increasingly popular imaging modality used in radiotherapy for patient setup and monitoring. It provides real-time feedback of the patient's position with respect to a reference surface dictated by either the external body contour of the treatment planning CT, or a surface capture acquired with the surface imaging system cameras. At the time of treatment, the patient's surface in the room is read by an optical system and automatically registered to the reference surface to calculate the deviation between the real-time and expected treatment positions using six degrees of freedom (6DOF). This information can then be used to evaluate and readjust the patient's setup from within the room without the use of ionizing radiation. More detailed descriptions of existing surface imaging systems can be found elsewhere. 1 Although trends may soon be changing in favor of eliminating the placement of skin marks, this tool is currently often still utilized in conjunction with laser alignment to tattoos. Radiographic imaging for image guided radiotherapy (IGRT) is still performed to ensure the precision of treatment delivery based on internal anatomy. 2 IGRT is an essential component of SBRT which employs immobilization devices and image localization techniques to treat small targets using hypofractionated dose regimens and millimeter PTV margins. 3 In the absence of optical surface imaging, it is common to initially position the patient based on skin marks and lasers, use orthogonal kV images to check overall alignment and match bony anatomy or fiducial markers, and finally refine target localization based on volumetric information from a cone beam CT (CBCT) scan. 4 To streamline the process, some centers bypass orthogonal imaging before CBCT. While this can be efficient if the initial patient position is adequate, it can also lead to increased patient imaging dose and extended setup time if alignment discrepancies, such as hip rotations or mispositioned extremities, cannot be corrected with automated couch movements and require re-acquisition of the CBCT to confirm satisfactory alignment prior to treatment. With surface guided radiotherapy (SGRT), positioning can be refined based on real time feedback during initial in-room setup, providing therapists the capability of detecting and correcting possible rotations or large translational discrepancies before leaving the room to acquire the CBCT. It is clear that SGRT cannot replace internal imaging for SBRT, but quantifying the effects of adding SGRT to the traditional IGRT chain for SBRT (referred to as SG-SBRT for the remainder of the text) can help elucidate the benefits of this technology. There is literature describing the benefits of utilizing SGRT for deep inspiration breath-hold treatments of left-sided breast cancer patients, [5][6][7][8][9] other breast cancer treatments, [10][11][12][13][14][15] and stereotactic radiosurgery, [16][17][18][19] but limited publications on its use for other sites or for initial positioning of SBRT patients. 20,21 The aim of this retrospective study is to establish the utility of optical surface imaging for initial patient setup in SBRT treatments and to formulate a proposed initial positioning process by studying the impact of orthogonal kV imaging when SG-SBRT is used and the effects of reference surface type selection (from treatment planning CT versus camera-acquired in the room) on its performance.

2.A | Patient selection and simulation
The use of patient data was reviewed by the Virginia Common- with standard definition cameras. This system consists of three pods, with two cameras each. Each pod also contains a projector that emits a pseudo-random speckled pattern of red light. The system uses this pattern to reconstruct the topography of the patient or object in its field of view. The resulting surface is then rigidly aligned to a reference surface based on a user-defined region of interest (ROI). For a more extensive description of the system, refer to the literature. 1 Patient data were divided into two cohorts based on whether or not surface imaging guidance was included in their treatment. The non-SG-SBRT group, treated in 2015 prior to clinical implementation of AlignRT, includes 37 patients (171 fractions). The SG-SBRT group consists of 26 patients (113 fractions) treated in 2016. All treatment courses ranged from 3 to 5 fractions, and treatment sites included primary and metastatic cancers of the lung, liver, spine, pancreas, and lymph nodes. Table 1 summarizes the information of the patient treatments included in this study. Some treatments were planned and delivered with the use of an active breathing coordinator (ABC) (Elekta Limited, Crawley, UK). For those patients, the planning CTs were obtained during inspiration breath hold. Patients treated in free breathing were simulated using a 4DCT scan with the 30% phase of the scan used as the primary image set to represent the mid-ventilation position. Spine patients were simulated in free breathing since respiratory motion does not affect the location of the target. Planning CT scans were acquired with a Philips Big Bore (Philips, Amsterdam, Netherlands) and the techniques used varied between 120 and 140 kVp depending on the patient's size and treatment site and 280 mAs for standard simulations and 600 mAs for 4DCTs, with 3 mm slice thickness for all scans, except spine (1.5 mm). The patients included in this study were planned in Pinnacle (Philips Radiation Oncology Systems, Fitchburg, Wisconsin, Version 9.6) and all external body contours were automatically created by using an outside-patient air threshold of 0.6 g/cm 3 . This structure, along with the treatment plan information, was sent to AlignRT using the RTPLAN and RTSTRUCT DICOM files.

2.B | Patient setup workflows
We investigated the differences between two patient setup workflows: the original procedure (non SG-SBRT) and the new one (SG-SBRT). The original clinical workflow involved laser alignment to patient skin marks, with couch shifts to the treatment isocenter if needed, followed by orthogonal kV imaging to correct for translational and rotational setup deviations with respect to bony anatomy. A CBCT was obtained for final target localization prior to the delivery of SBRT. In the new workflow, AlignRT was introduced after laser alignment and before kV imaging to refine the patient's position in the room (Fig. 1). Although skin mark alignment can be replaced with SG, this step was kept in the new workflow for easier implementation of surface imaging as therapists were still growing accustomed to the system. When using AlignRT for initial setup, an in-room monitor displays the adjustments needed to correct the patient position in real-time using a continuous feedback loop. These adjustments are given as three translational (vertical, longitudinal, lateral) and three rotational (yaw, roll, pitch) deltas based on an automatic rigid registration between the real-time surface of the patient in the room and the selected reference. As mentioned in the introduction, the reference surface can be based on the external body contour of the planning CT (DICOM reference), or acquired using the in-room optical cameras (VRT reference). The registration only focuses on the area encompassed by the user-defined region of interest (ROI). Both the accuracy and refresh framerate of the deltas depend on the ROI used for registration. In our workflow, the DICOM reference was always used for initial positioning throughout the treatment, and the ROI was defined following vendor recommendations for different treatment sites (Fig. 2). Prior to clinical use each day, therapists were instructed to perform the vendor-recommended daily test verification on the surface imaging system to ensure performance was satisfactory (root-mean-square position of the isocenter as measured by the system was within 1mm of calibration). If this test showed a deviation beyond the expected value, the system was recalibrated.
When positioning patients, therapists were asked to achieve delta values as close to zero as possible before completing the in-room portion of the setup phase. After refining the patient position with surface imaging, the remaining setup proceeded as usual, with orthogonal kV imaging followed by CBCT. After the treatment position was confirmed based on CBCT, a VRT reference image was captured for intrafraction treatment monitoring.
For either workflow, due to the lack of 6DOF capabilities of the treatment couch, therapists manually adjusted the patient to correct rotational discrepancies deemed large enough to affect the quality of the treatment delivery upon CBCT inspection by the physicist and attending physician. Any time a rotational modification was performed, a second CBCT scan was then acquired to verify the adjusted patient position. Per department policy, an additional CBCT scan is also required to confirm the patient's alignment prior to treatment if translational shifts on imaging are found to be larger than 8mm in any direction, or 15mm when the absolute value of all three translational shifts are summed. This policy is in place to ensure that the patient's position is still satisfactory before starting treatment after large imaging shifts have been applied.

2.C | Data analysis
The 284 SBRT fractions in this analysis included 171 fractions treated prior to the clinical implementation of surface imaging (original workflow, non-SG-SBRT), and 113 treated with the inclusion of AlignRT (SG-SBRT workflow).
To assess the impact of SG on initial setup, we compared the difference in the initial couch position after in-room alignment but before kV imaging, to the treatment couch position after final target localization using CBCT, for setups with and without surface imaging. The shifts applied based on every image registration at the treatment machine are automatically saved with each image in the record and verify system, Aria (Varian, Palo Alto, California, Version 11).
Hence, these provide the difference in the couch position before and after imaging. The absolute value of these differences was used in the analysis. Although this is a simple analysis, with limitations that will be discussed in a latter section, it is a useful quantitative metric to compare in-room positioning performance of AlignRT versus laser localization alone. The time difference between the two workflows could not be quantified since the record and verify system has no way to track the time taken to perform laser localization or surface imaging adjustments. The necessity of orthogonal kV The differences in registration between the VRT surfaces captured at each fraction were evaluated to assess day-to-day surface reproducibility, both against the DICOM reference surface, as well as to the VRT surface captured at each previously treated fraction The p-values of the differences between the two data sets for all parameters studied were calculated using a Wilcoxon Rank-Sum test (α = 0.05) since the data are not normally distributed.

| RESULTS
A comparison of the initial couch position at the start of kV imaging to final couch position after CBCT imaging demonstrates a smaller range and median deviation in all three translational directions and vector magnitude when optical surface imaging is included in the workflow. Table 2  group where the translational deviation was greater than 1cm. There was also a statistically significant difference in the rotations found amongst the two groups. Table 2 also shows the median, quartile 1, quartile 3, and maximum rotations along the three directions (pitch, yaw, roll). Overall, the rotations for the SG-SBRT group were smaller in a statistically significant manner, although the maximum roll value calculated for that group was slightly larger than that of the non-SG-SBRT arm. Figure 4 (chest or abdomen) that benefited more from the addition of orthogonal kV imaging, the results did not show a difference (see Table 3). Not surprisingly, the addition of orthogonal kV imaging is beneficial for bony targets in the non-SG-SBRT group. The difference in the average number of CBCTs between the two groups is not statistically significant.
Of the 113 patients in the SG-SBRT group, 102 had VRT reference captures.  3. Comparison of differences in daily VRT reference captures compared to 1) the DICOM reference surface (top) and 2) the VRT reference capture from the previously treated fraction.
T A B L E 2 Couch position differences from pre-orthogonal kV imaging to post-CBCT localization, for the non-SG-SBRT and SG-SBRT groups. they provide a more comprehensive and robust method of assessing the patient's position relative to the plan. Since some surface imaging commercial systems require users to manually select the region of interest, the quality of the setup achieved based on surface imaging will depend on appropriate ROI definition and user proficiency.
Nevertheless, the data also demonstrates that surface imaging helps detect large setup errors prior to imaging (see Table 2). For the non-SG-SBRT group, 2.9% of fractions had translational differences greater than 3 cm, 5.8% had greater than 2 cm, and 45.6% greater than 1 cm. The SG-SBRT group did not have any fractions with discrepancies larger than 3 cm, had 1.8% with differences greater than  the study being retrospective, it is also impossible to acquire time estimates of how long therapists spent during initial setup performing laser localization with skin marks alone versus SG prior to imaging. This information would also be valuable in comparing the two processes, since if one is considerably longer than the other, assuming the comfort level of the users is the same for both, this should be factored into the evaluation as it can affect clinical workflow and patient comfort.
While the reference surface study found that the mean devia-  As this is an offline retrospective analysis, the surface study is a simplified representation of the potential differences between the DICOM and VRT reference surfaces. It does not include data from F I G . 5. Illustration of the potential propagation of systematic errors when performing initial setup for a patient based on a surface imaging system acquired daily reference surface.
actual patient setup to daily reference captures, so it lacks the variations that the manual alignment would introduce when therapists process the feedback from the system in real time as they are adjusting the patient's position. To truly assess whether such an effect exists, a prospective trial randomizing patients to daily setup after the first fraction using the DICOM versus the previously acquired VRT reference surface must be conducted to determine whether there is a statistically significant difference in shifts between the two methods.
Lastly, the data collected from the treatments delivered with AlignRT began when the system was newly implemented and thus, presents a gradual refinement in procedural definition and proficiency of therapists in its use. While familiarity with the system gradually improved over time, AlignRT was installed on a single TrueBeam linear accelerator at our institution. As a result, proficiency in setup with the system was not uniform among radiation therapists, who routinely rotate assignments between treatment machines. Thus, the patient setups with AlignRT may not demonstrate maximum proficiency attained in comparison to setup with laser localization, which is a standardized skillset among all radiation therapists at our institution. Despite this, the SG-SBRT cohort still shows a statistically significant decrease in positional discrepancies when compared to the non-SG-SBRT group.

| CONCLUSIONS
The addition of surface imaging was found to improve the precision and safety of initial patient setup for SBRT treatments. Patients in the SG-SBRT cohort had translations and rotations between the initial position and treatment position that were statistically significantly less than the non-SG-SBRT cohort. Although the number of CBCTs acquired for setup was similar in both groups, the addition of orthogonal kV imaging to the initial setup process was only valuable to keep the CBCT shifts below re-imaging thresholds for less than 19% of the SG-SBRT fractions compared to 51% of the non-SG-SBRT ones.
Hence, the inclusion of orthogonal kV imaging as part of the initial setup process could be re-evaluated for SBRT patients when using surface image guidance. The choice of reference surface for initial positioning with surface imaging does not make a statistically significant difference in the outcome; however care should be taken to avoid systematic propagation of positional discrepancies when using a camera-acquired instead of a DICOM reference.

CONF LICT OF I NTEREST
Dr. Padilla reports nonfinancial support from Vision RT, during the conduct of the study; nonfinancial support from Vision RT and a grant from Philips, outside the submitted work. Mr. Leong has nothing to disclose.