Single‐institution report of setup margins of voluntary deep‐inspiration breath‐hold (DIBH) whole breast radiotherapy implemented with real‐time surface imaging

Abstract Purpose We calculated setup margins for whole breast radiotherapy during voluntary deep‐inspiration breath‐hold (vDIBH) using real‐time surface imaging (SI). Methods and Materials Patients (n = 58) with a 27‐to‐31 split between right‐ and left‐sided cancers were analyzed. Treatment beams were gated using AlignRT by registering the whole breast region‐of‐interest to the surface generated from the simulation CT scan. AlignRT recorded (three‐dimensional) 3D displacements and the beam‐on‐state every 0.3 s. Means and standard deviations of the displacements during vDIBH for each fraction were used to calculate setup margins. Intra‐DIBH stability and the intrafraction reproducibility were estimated from the medians of the 5th to 95th percentile range of the translations in each breath‐hold and fraction, respectively. Results A total of 7269 breath‐holds were detected over 1305 fractions in which a median dose of 200 cGy was delivered. Each fraction was monitored for 5.95 ± 2.44 min. Calculated setup margins were 4.8 mm (A/P), 4.9 mm (S/I), and 6.4 mm (L/R). The intra‐DIBH stability and the intrafraction reproducibility were ≤0.7 mm and ≤2.2 mm, respectively. The isotropic margin according to SI (9.2 mm) was comparable to other institutions’ calculations that relied on x‐ray imaging and/or spirometry for patients with left‐sided cancer (9.8–11.0 mm). Likewise, intra‐DIBH variability and intrafraction reproducibility of breast surface measured with SI agreed with spirometry‐based positioning to within 1.2 and 0.36 mm, respectively. Conclusions We demonstrated that intra‐DIBH variability, intrafraction reproducibility, and setup margins are similar to those reported by peer studies who utilized spirometry‐based positioning.


| INTRODUCTION
Adjuvant whole breast radiotherapy (WBRT) following lumpectomy improves local control and in some populations overall survival in the treatment of invasive breast cancer. 1,2 However, collateral toxicities to the heart and lungs remain a significant challenge. 3,4 Deep-inspiration breath-hold (DIBH) harnesses the advantage of organ motion during the respiratory cycle to minimize overlap of the heart and lungs with the treatment fields. 5 Compared to free-breathing, DIBH significantly reduces the dose to the heart for left-sided treatments [5][6][7] and the ipsilateral lung for right-sided treatments. 8 To minimize irradiation of nearby organs-at-risk without sacrificing treatment of the target volume, daily patient setup and breath-hold reproducibility, which can be monitored through cine portal imaging, spirometry or surface imaging (SI), are essential. Even with spirometric-DIBH as an alternative to voluntary DIBH (vDIBH), breast surface reproducibility varies 1-3 mm. 9 Surface imaging has the added benefit of being noninvasive and voluntary.
Image guidance is frequently used to detect large setup errors and refine patient positioning in the setting of radiotherapy. 10 Modalities such as electronic portal imaging devices, cone-beam CT (CBCT), and surface imaging have been implemented to verify setup accuracy of DIBH patients. [11][12][13][14][15][16] However, daily shifts in patient setup are inevitable even with image guidance. To buffer these daily setup errors and ensure adequate dosing to target tissue, the planning treatment volume (PTV) incorporates margins to the clinical target volume (CTV) taking into consideration the systematic and random errors of daily setup. 17 We refer to PTV margins as setup margins, because these calculations serve as a measure of setup reproducibility in this study rather than as values incorporated clinically into radiotherapy plans.
Unlike radiation-based image guidance systems that measure displacements in bony anatomy, surface imaging provides real-time tracking of breast position without dosimetric detriment to the patient. To our knowledge, we are the first to calculate setup margins achieved by vDIBH with real-time surface imaging of whole breast targets, and compare these values to those achieved in peer studies that use spirometric-DIBH and/or x ray-based monitoring.
Use of a reference surface from the simulation CT scan throughout the entire treatment course allows us to report on systematic and random errors of the breast surface during vDIBH calculated using real-time surface imaging data.  (Smithers Med-ical Systems, Canton, OH) with or without the SaBella Flex breastboard (CDR Systems, Calgary, Alberta, CAN), depending on whether it had been implemented. Molds were designed to encompass the upper body, arms, and hands, which were positioned actively (i.e., by gripping pegs) or passively above the head. Therapists coached the patients over three to five breath-holds to achieve a consistent maximum inhalation position that could be maintained for 15 s whose amplitudes were monitored using the respiratory motion assessment (RPM, Varian Medical Systems, Palo Alto, CA) device placed at midline just above the xiphoid. Two sequential CT scans were acquired on a Brilliance BigBore scanner (Phillips Healthcare, Andover, MA) with 3-mm slice thicknesses during free-breathing (FB) and vDIBH.

2.A | Patient selection and simulation
FB scans were used for dosimetric comparisons and to generate a reference surface for initial treatment positioning. DIBH CT scans were used for treatment planning and generation of a reference surface for breath-hold guidance. Reference surfaces were automatically contoured in Pinnacle v9.0-9.6 (Philips Systems, Andover, MA) using a CT density threshold of 0.6 g/cm 3 . The reference surface was exported to AlignRT as a DICOM structure file. AlignRT performs some down-sampling of the 3D vertices used to define the surface as a triangle mesh surface, whose resolution depends upon the anatomical site chosen within the system and the geometry of the surface.

2.B | Treatment planning
Patients were treated with 6-MV or 15-MV photons on a Varian Trilogy linear accelerator with an OBI console (Varian Medical Systems, Palo Alto, CA). Tangential fields (i.e., 2-field) were optimized with the field-in-field technique to improve dose homogeneity throughout the breast. Anterior and posterior oblique supraclavicular fields (i.e., 3-field) using a mono-isocentric technique were added to treat nodal targets when deemed appropriate. Dose calculations with correction for tissue heterogeneities were performed in Pinnacle v9.0-9.6 with the aim of covering the target, with at least 95% of the prescribed dose while minimizing 20 Gy, 30 Gy, and mean doses to the heart and lungs when compared to FB plans per institutional guidelines. 18 Breast and nodal targets were contoured per the RTOG Breast Cancer Atlas.

2.C | Surface imaging
The AlignRT v5.0 three-camera system (VisionRT, London, UK) was used for initial positioning and during treatment to continuously monitor surface displacements in three translational dimensions (A/P: anterior-posterior; L/R: left-right; and S/I: superior-inferior) and in three rotational dimensions (yaw, roll, and pitch). Displacements are calculated automatically by the software following rigid registration using a proprietary iterative closest point algorithm of a single region-of-interest (ROI) delineated by the user on the reference surface. The "Entire" ROI including arms and chin was selected for initial FB positioning while the "Breast" ROI mimicking the projection of the tangential fields was selected for treatment verification at the DIBH position. 19 The workflow for patient positioning is depicted in Fig. 1(a). Initial positioning began by aligning skin tattoos to in-room lasers followed by minimizing the displacements between the real-time images and the FB reference surface within the "Entire" ROI.
Patients were then instructed to hold their breath and coached or adjusted until real-time displacements of the 'Breast' ROI agreed with the DIBH reference surface to within 3 mm/1°in each dimension. Beam gating was enabled during treatment using the tightest 3D thresholds to within a range of 5-7 mm and 2-3°, as determined from the patient's DIBH reproducibility during the first treatment.
X-ray imaging was performed once pretreatment, during the first treatment, and weekly thereafter using an orthogonal kV pair and MV ports, which were used to correct the patient's position as needed. Figure 1(  script was developed to process these displacements which resulted from registration to the DIBH reference surface. Individual breathholds were identified as periods when the radiation beam state was "ON," which was triggered by the requirement that the patient surface be within tolerance of the reference surface (Fig. 2). Consecutive breath-holds were distinguished by a temporal separation of at least 5 s, whose value stemmed from knowledge of our clinical workflow.

2.E | Statistical analysis
Since setup margins cannot be statistically compared, 21 Figure 3 shows a histogram distribution of mean dimensional displacements for all fractions, which shows a mean translational shift of >1 mm in the A/P direction that does not exist for S/I or R/L. This may be due to a systematic discrepancy between the CT-generated surface and the true breast surface in the A/P direction. van Herk's recipe reveals that a uniform 7-mm margin would encompass greater than 95% of daily shifts, although stricter 5-mm margins would suffice for A/P and S/I dimensions. In the subgroup analysis, only the mean error in S/I between 2-field and 3-field treatments differed (P < 0.0167) indicating that systematic S/I errors were significantly smaller for 3-field treatments. There were no statistically Comparisons of isotropic setup margins and standard deviations of the systematic and random errors are shown in Table 3 between the left-sided breast group in this study and those in Conroy et al. 14 and Yang et al. 15 , which both use x-ray monitoring of left-sided DIBH. The use of surface imaging in this study achieved a comparable setup margin to those in aforementioned peer works.

| DISCUSSION
The DIBH techniques offer proven advantages in breast radiotherapy via dosimetric sparing of organs-at-risk. 5 are comparable to our results in A/P (4.9 mm) and S/I (4.7 mm) but not in L/R (6.2 mm). This discrepancy could be because a singlecamera AlignRT system was used in the Alderliesten study resulting in truncation of the "Breast" surface, or due to differences in patient habitus between the two study populations as two-thirds of patients in this study were overweight or obese (Table 1). Isotropic setup margins were also calculated ( postulated that this was related to less tissue deformation around the isocenter location for 3-field treatments compared to 2-field treatments, whose isocenters were located in deformable breast tissue. In our study, all patients were able to perform vDIBH and only~5% required resimulation due to either anatomical or breathhold reproducibility changes.
While it might be assumed that spirometric control of DIBH results in exact breath-hold reproducibility over vDIBH, surface tracking of external infrared thoraco-abdominal markers on seven patients during spirometric-DIBH showed intrafraction variation in the breast surfaces on the order of 1.75-2.5 mm with lung volumes. 9 When comparing intra-DIBH and intrafraction variability in the breast surface achieved by vDIBH to Fassi et al. 9 , we found slightly better intra-DIBH stability in vDIBH over spirometric-DIBH, and comparable intrafraction reproducibility regardless of breathhold technique indicating that vDIBH measures up to spirometry.
The implication of this finding is tremendous given that vDIBH is far less invasive, and is vastly preferred to spirometry by both patients and therapists. 14,22 While comparison with Fassi et al. 9 invites the possibility that vDIBH may offer an advantage with respect to intra-DIBH stability, Table 4 showed that a 5-mm threshold is still necessary to account for 95% of intrafraction variability in each dimension, which corresponds to a 7-mm threshold on the 3D magnitude vector as required by the AlignRT v5.0 software. This is a larger threshold than the F I G . 3. Distribution of translational displacements across all treatment fractions (n = 1305). All three dimensions are overlaid using bars of increasing width, with the thinnest for A/P and the widest for R/L. One limitation of this study is that the reference surface was generated from CT scan data, which may not perfectly match the surface rendered by the SI cameras. A histogram of the distributions of translational displacements (Fig. 3) shows that the distribution of A/P displacements is offset by just over 1 mm, whereas the distribution of S/I and L/R displacements are roughly centered on zero. The shift in distribution of A/P displacements, which has been corroborated by others, 13 may stem from a systematic bias resulting from the autosegmentation of the CT external surface using a specific CT density. Li et al. observed a larger bias in the A/P dimension than in the other dimensions, which could be as large as 0.8 mm depending on the CT density used to generate the reference surface. 30 If this were true, the translational shifts reported on in this study would represent an upper limit and in fact, setup reproducibility could be better than the estimates we provide here.
Another limitation is that daily discrepancies from the CT-simulated position may result in dosimetric deviations from the intended treatment plan.

| CONCLUSION S
In this study, we report on our institution's setup margins and DIBH stability and reproducibility achieved by real-time surface imaging of T A B L E 4 Intra-DIBH stability and intrafraction reproducibility from 31 left-sided patients in this analysis compared to those of spirometrybased results from seven patients as reported on in Fassi et al.

CONF LICT OF I NTEREST
No conflicts exist.