Accuracy of surface‐guided patient setup for conventional radiotherapy of brain and nasopharynx cancer

Abstract Purpose To evaluate the accuracy of surface‐guided radiotherapy (SGRT) in cranial patient setup by direct comparison between optical surface imaging (OSI) and cone‐beam computed tomography (CBCT), before applying SGRT‐only setup for conventional radiotherapy of brain and nasopharynx cancer. Methods and Materials Using CBCT as reference, SGRT setup accuracy was examined based on 269 patients (415 treatments) treated with frameless cranial stereotactic radiosurgery (SRS) during 2018‐2019. Patients were immobilized in customized head molds and open‐face masks and monitored using OSI during treatment. The facial skin area in planning CT was used as OSI region of interest (ROI) for automatic surface alignment and the skull was used as the landmark for automatic CBCT/CT registration. A 6 degrees of freedom (6DOF) couch was used. Immediately after CBCT setup, an OSI verification image was captured, recording the SGRT setup differences. These differences were analyzed in 6DOFs and as a function of isocenter positions away from the anterior surface to assess OSI‐ROI bias. The SGRT in‐room setup time was estimated and compared with CBCT and orthogonal 2D kilovoltage (2DkV) setups. Results The SGRT setup difference (magnitude) is found to be 1.0 ± 2.5 mm and 0.1˚±1.4˚ on average among 415 treatments and within 5 mm/3˚ with greater than 95% confidence level (P < 0.001). Outliers were observed for very‐posterior isocenters: 15 differences (3.6%) are >5.0mm and 9 (2.2%) are >3.0˚. The setup differences show minor correlations (|r| < 0.45) between translational and rotational DOFs and a minor increasing trend (<1.0 mm) in the anterior‐to‐posterior direction. The SGRT setup time is 0.8 ± 0.3 min, much shorter than CBCT (5 ± 2 min) and 2DkV (2 ± 1 min) setups. Conclusion This study demonstrates that SGRT has sufficient accuracy for fast in‐room patient setup and allows real‐time motion monitoring for beam holding during treatment, potentially useful to guide radiotherapy of brain and nasopharynx cancer with standard fractionation.


| INTRODUCTION
Surface-guided radiotherapy (SGRT) using optical surface imaging (OSI), as a special form of image-guided radiotherapy (IGRT), has been increasingly applied to guide patient setup and monitor patient motion during treatments, such as cranial frameless stereotactic radiosurgery (SRS) and left-sided breast deep-inspiration breath-hold (DIBH) treatment. [1][2][3] The advantages of SGRT include nonionization radiation 3D imaging with the patient's external anatomy and realtime 4D imaging for motion tracking and threshold gating. Most clinical applications involve patient setup and motion monitoring either for rigid anatomy with a fixed relationship between the skin surface and deep-seated lesions, such as brain cancer, [4][5][6][7] or for superficial lesions, such as breast cancer. [8][9][10][11] For patient setup, the external body contour from the planning CT is mostly used as the reference, while for motion monitoring, an on-site surface reference is usually captured, excluding the residual setup error. The OSI has recently been applied to achieve tattoo-free SGRT patient setup and motion monitoring as a replacement of conventional tattoo-laser alignment setup. 10,11 Other clinical applications may include SGRT for deformable anatomy with external-internal motion modelling, [12][13][14] patientgantry collision detection during radiotherapy, 15,16 and patient identification and registration via facial recognition. 17 For brain and head-and-neck (HN) patients treated with conventional fractionation, room lasers are often used to align the patient with native anatomic landmarks, such as the nose, eyes, and tragus, as well as the tattoos, bb's, or cast lines on the thermoplastic masks. 18 Due to the large setup uncertainties, the safety margin for partial brain and HN setups are usually 5-6 mm based on 2 CBCT studies 19,20 and 3-10 mm for conventional setup and 3-6 mm using helical tomotherapy CT for setup in another study. 21 Wang et al. studied 22 patients with 505 CBCT setups and concluded 5-6 mm safety margin for conventional setup and 3 mm safety margin for CBCT setup. 19 Leitzen et al. studied 15 HN patients using megavoltage CT (MVCT) for IGRT setup and evaluated the setup margin of 3-5 mm for MVCT-based setup and 3-10 mm for conventional setup. 21 Gopan and Wu evaluated SGRT setup accuracy in 11 HN patients with simulated OSI patient surface contours from 77 helical computed tomography (CT) images including planning CT during their 6-week treatment courses. 22 They found that the setup uncertainty less than 5 mm had 90% confidence level and higher setup uncertainties occur as the site is further away from the skull due to the deformable cervical spine. However, as the study does not involve OSI imaging, it was not under clinical conditions as any interference items, such as facial masks, were excluded. Kuo et al.
reported a phantom study using CBCT, on-board and ceiling-floormounted 2DkV, as well as OSI based on both anatomic-based and point-based registration. 23 Both isocenter and target registration errors were reported and a maximum of 2.5 mm OSI uncertainty was found. To apply SGRT-only patient setup in the clinic to treat brain and HN patient with standard fractionation, a thorough evaluation of SGRT setup accuracy and other advantages is needed with a large clinical dataset.
In this study, we investigated 415 treatments of 269 brain patients who were set up with both OSI and cone-beam CT (CBCT) for the SRS treatments, and the SGRT setup differences were acquired prospectively in 2018-2019 and evaluated retrospectively in 6 degrees of freedom (DOF) using CBCT as the reference. In our clinical protocol, we specifically requested to capture an OSI verification image immediately after CBCT setup, so that it was possible to assess SGRT setup differences through direct comparison. Due to registration landmark differences in these two imaging modalities, the SGRT setup differences were evaluated as a function of the distance of the isocenter to the skin surface, in addition to generic statistical analyses. The objective of this study is to assess the feasibility, accuracy, and time requirement of SGRT-only setup without skin markers, daily 2DkV, or CBCT for radiotherapy of brain or nasopharynx cancer with the standard fractionation.

| METHODS AND MATERIALS
In this study, 415 treatment setups of 269 patients with both OSI and CBCT were analyzed. A treatment protocol was designed with initial SGRT, then CBCT setup, followed immediately by capturing a verification AlignRT image, illustrating the difference between OSI and CBCT. Some earlier-treated patients also received 2DkV verification until this step was removed from the protocol. These patients were treated with brain SRS from 2018 to 2019 using a CDR head immobilization device on the CDR couch extension (CDR, Calgary, Canada) attached to a PerfectPitch couch in 6 degrees of freedom (DOF) of a TrueBeam machine with HD multileaf collimators (Varian, Palo Alto, CA). Before the patient entered the room, the treatment isocenter was determined based on 3 ball-bearing (BB's) on the open-face mask and the setup instruction with the isocenter center shifts from the reference point using the room lasers. Then, the couch position was acquired to facilitate the setup. The OSI system was AlignRT (VisionRT, London, UK), which was used for initial SGRT  within the open-face mask area above the lips on the external contour of the planning CT. The goal is to use the SRS patient dataset to estimate SGRT-only setup accuracy for radiotherapy of brain and nasopharynx cancer with conventional fractionation.

2.A | Description of patient data acquired from SGRT and IGRT stereotactic radiosurgery
In the SRS treatment procedure, all patients were initially set up inside the room with 6DOF SGRT guidance in real-time delta (RTD) mode. After a patient was positioned into the CDR immobilization device, the patient's head rotation was first corrected by adjusting head position and then the Pitch/Roll knobs of the CDR couch extension after the open-face mask placement, followed by translational correction with couch shift. Before closing the door for CBCT, the RTD was turned on to monitor patient motion during CBCT, CBCT/CT image registration, and setup verification from a physicist and approval by a physician. This process also warmed up the AlignRT system in the RTD mode to eliminate the baseline-drift error, 3,5,7 which produces an error of~0.3 mm due to the thermal heat in the camera systems but stabilized after 5-to 10-minute RTD.
After CBCT/CT registration in 6DOF, the shifts were applied from the console using the 6DOF couch. Because of the use of initial SGRT setup, CBCT shifts were small, usually within 2 mm and 1i n any DOF. Immediately after CBCT shifts were applied, an OSI verification image was captured, recording the difference between OSI and CBCT. These differences for all 415 treatments (269 patients) were used to evaluate SGRT-only setup accuracy. Then, an on-site OSI reference image was captured for motion monitoring during treatment. The time duration of SGRT setup was saved in the patient's data folder and the CBCT/2DkV SRS patient setups and treatments were recorded in the ARIA Offline Review (Varian, Palo Alto, CA). The data were used to estimate the time spending and saving for the SGRT-only setup.

2.B | SGRT setup uncertainty assessment and statistical analyses
Various statistical analyses were performed, including SGRT setupdifference distribution and correlation among the 6DOFs. Significance level (α) of 0.05 was used for rejecting the null hypothesis and was adjusted for multiple comparisons when applicable using the Bonferroni method. 24 Statistical significance of difference among group means was tested using a Student's t-test for 2 groups and Analysis of Variance for more than 2 groups. These tests were performed using R version 3.6.1. Correlation between translational and rotational shifts was assessed based on the correlation calculation among the 6DOF.
The dependency of SGRT setup differences on the isocenter location was assessed because the OSI ROI had a bias on the anterior surface, unlike the skull landmark that "evenly" spreads in all directions of the head in CBCT-to-planning-CT registration. The location of the registration landmark was a major difference between the OSI and CBCT. Therefore, we hypothesize that there is a dependency of SGRT setup differences with the isocenter vertical location: the farther away from the anterior ROI, the larger the setup difference would become.
For simplicity, the 6DOF residual OSI differences were summarized into 2 translational and rotational components by taking the magnitude (MAG) of the translational and rotational vectors with 3 components: where AP is the anterior-posterior (or vertical, VRT) direction and axis for the yaw rotation, SI is the superior-inferior (or longitudinal, LNG) direction and axis for the roll rotation, and LR is the left-right (or lateral, LAT) direction and axis for the pitch rotation. Note that the MAG rot is purely a vector magnitude, similar to MAG trans , but does not have associated physical meaning. The location of the isocenter was categorized in two ways: First, using the brainstem as a reference, 3 zones were defined in the AP and LR directions and 2 zones in the SI direction. Second, the plans were grouped into four ranges of skin-to-isocenter distance, which was defined by projecting the isocenter to the midplane in the LR direction (laterally "centralized" isocenter) and measuring the distance to the anterior surface along the lateral midplane. Using these 2 methods, the anterior ROI was assessed for potential setup bias in any directions.  3.B | SGRT setup differences in relationship with the vertical distance of "centralized" isocenter Table 2  There was a noticeable difference in the pitch and SI setup differences, but not statistically significant due to a small number of incidences. It is worthwhile to mention that there is a large difference in the pitch setup, suggesting that the head "nodding" motion may be the primary cause of patient motion during treatment. These motion outliers occur infrequently with relatively small motion (<2 mm beyond the SRS tolerance).  In addition to many other differences between OSI and CBCT, the location of their ROI is a major assessable difference. Because SGRT surface registration is based on the ROI, which is on the anterior facial surface of the patient's head, there is a bias toward the anterior anatomy, unlike the skull landmark in CBCT-to-planning-CT registration. SGRT only aligns to the partial anterior surface of the head and the posterior alignment is unknown. This triggered us to make the hypothesis that the SGRT setup difference is dependent of the isocenter, meaning the farther away from the anterior surface ROI, the larger the setup difference. In fact, this study has illustrated that the SGRT setup difference is a function of the vertical skin-toisocenter distance. The results in Table 3 and Fig. 2 suggest that the SGRT setup difference is linearly increasing as the vertical distance increases. However, the differences between most anterior and most posterior zones are mild, less than 1.0 mm, implying that the anterior bias of the ROI is unlikely to be as clinically impactful as initially thought. Other than the AP direction, we did not find any significant trend of dependency on the isocenter location.

3.D | SGRT-only setup time comparing with CBCT and 2DkV setup times
It is worthwhile to mention that a mild negative correlation was observed between SI translation and pitch rotation (r = −0.29) and between LR translation and roll rotation (correlation = −0.44), suggesting possible ambiguity in surface registration results. This means that a translation shift may be replaced by the corresponding rotational shift, resulting in almost equally well-registered surfaces. Interestingly, a mild correlation (r = 0.37) between Yaw and Roll rotations is also shown in Fig. 1. In the clinic, such phenomena may have been observed and this study provides the quantitative analysis of the observation.
T A B L E 2 Relationships between the mean SGRT translational/ rotational difference from CBCT and isocenter location. The magnitude of the translational (Δ trans ) and rotational (Δ rot ) differences were used [defined in Eqs. (1) and (2)].

Isocenter location
Translation difference (mm) Bold emphasizes the P-value. a The P-value is calculated using a two-sample t-test (SI) or analysis of variance (AP. LR). b The brainstem is used as the reference to separate the brain into different sections.
F I G . 2. Boxplots for distributions of translational and rotational differences in isocenter zones and anterior SID (skin-to-isocenter distance). The median and 25%-75% percentile are shown in a box, together with outliers (dots), which are associated with very large SID, away from the anterior region of interest (ROI).
T A B L E 3 Relationships between the mean SGRT translational/ rotational differences and the vertical depth from isocenter to the anterior surface at the midline of the brain. The magnitude of the translational (Δ trans ) and rotational (Δ rot ) differences were used [defined in Eqs. (1) and (2)]. Bold emphasizes the P-value. a The vertical skin-to-isocenter distance is obtained by first shifting the isocenter laterally to the midline. This further analysis confirms the initial results in Table 2. b The P-value is calculated for differences in means across four depth ranges (Q1 to Q4) using the Analysis of Variance.

4.B | Clinical benefits for SGRT-only setup and option for motion monitoring
In IGRT patient setup using CBCT and 2DkV, it requires staff to leave the room and close the door, making it impossible to adjust the patient position. Although CBCT and 2DkV align with the internal bony structures, they do give the patient extra radiation dose, take longer time to scan, and require physician for approval before treatment. In contrast, SGRT patient setup is performed inside the room, the surface image is registered automatically, and the setup does not require the physician to approve, therefore, the treatment may start as soon as the automatic alignment meets the clinical criteria. However, as demonstrated in this study, the SGRT setup uncertainty has a magnitude of 1.0 AE 2.5 mm and 0.1˚AE1.4˚and is within 5mm/3˚at >95% confidence level, and SGRT-only setup is usually less than 1 min to complete, reduced by one-order of magnitude. Compared with previously reported conventional setup accuracy, SGRT provides an advantage while the setup time is roughly the same.
Within the OSI ROI, the skin deformation is limited, unless a patient experiences substantial weight changes, including weight loss due to the disease or weight gain due to possible hormone therapy. The ROI is composed of many points in the highest resolution forming a surface mesh. Therefore, the 3D surface alignment is more reliable than 3 skin markers, tattoo points, or cast lines on the mask. In addition, the integrity of the 3D surface is more reliable, unlike skin markers/tattoos that can be moved around by  | 55 fractionation may not need SGRT motion monitoring if the SRS patient immobilization device is applied as the patient motion incident rate is low (<4%) and the magnitude of the motion is relatively small (<3 mm). 30 4.C | Other concerns, limitations, and future directions on SGRT When using SGRT-only patient setup and motion monitoring, the patient immobilization device is important, and here, the SGRT-only setup accuracy is derived from the CDR system with customized head mold and open-face mask. It has been reported that patient setup accuracy is higher using individual customized head support compared with standard headrest. 31 In addition, the customized head immobilization system also provides much higher patient motion restriction, reducing possible intrafractional patient motion.
In this study, we also see <4% outliers that have >5 mm SGRT setup differences due to the large vertical skin-to-isocenter distance of centralized isocenter. The outliers often have a large SI translational difference and large pitch rotational difference, as shown in Fig. 3. Interestingly, we also observed a mild correlation between the SI translation difference and pitch rotation difference (r = −0.29), suggesting that the 2 differences are related, namely raising the uncertainty in one would result in enlarged uncertainty in the other.
Similarly, we observed cases in which the isocenter is near the lateral brain edge and SGRT setups have large LR and Roll differences, which are also mildly correlated (r =−0.44). These cases attribute to the largest differences (outliers) as shown in Table 1 In our clinic, the initial efforts in patient data preparation are made by the dosimetrists: from a planning system to SGRT system and from isocenter check to ROI creation. However, to make the ROI truly patient specific and optimal for SGRT, therapists at the treatment console should be able to exclude deformable skin from the ROI with the guidance of a color-coded deformation tool. We have trained the therapists to modify the patient-specific ROI at the treatment for both SGRT patient setup and motion monitoring.
As a continuation to our efforts in patient setup and motion monitoring using SGRT only for radiotherapy of brain and nasopharynx cancer patients with conventional fractionations, we will put efforts to study other anatomical sites to go tattoo free, such as head and neck and breast treatments. For each anatomical site, there are site-specific concerns and the conclusion from a study on one site may not be directly applicable to another site without additional site-specific investigation.

| CONCLUSION
In this study, we investigated the SGRT-only setup accuracy using

CONFLI CT OF INTEREST
The authors have no relevant conflict of interest to disclose.

DATA SHARING
For patient setup data used in this study, we will make efforts for anonymization and make them available upon request from readers.

AUTHOR CONTRI BUTION
All authors have made significant contributions to this clinical study to be qualified as coauthors.