As a retrospective study approved by the IRB, the clinical experience with 48 patients treated at our institution (Beijing Cancer Hospital, China) is presented. Over the cohort, 30 patients were male and 18 were female. The patient ages ranged from 44 to 77 with the median as 63. The patients were with more than five brain metastases and received whole brain irradiation. CTV was the entire brain, and a 3-mm isotropic margin was added to define PTV. The prescription was 30 Gy /10 fractions. All of the patients were immobilized with open-face masks and set up using the proposed surface guided setup workflow. The entire clinical chain from simulation to treatment as well as data analysis is detailed as below.
Pretreatment
Patient immobilization
The open-face double shell positioning system (MacroMedics, Belgium) depicted in Fig. 1 was used at our institution for SRT patients. It consisted of two shells that were hosted on a dedicated support structure indexed onto the setup board. During mask molding, the anterior shell was left open to expose the face area, which included the nose, eyes and cheekbones. The exposed area can be used as ROI for AlignRT to track during optical surface imaging. The posterior shell, made of thick thermoplastic, was designed to offer a sturdy and personalized headrest that would improve patient comfort.
Patient simulation and planning
After the creation of open-face masks, patients were scanned on our CT-Sim (SOMATION Sensation Open, Siemens Healthineers, Germany). During simulation, four cross markers for localization were drawn on the tapes attached to the mask as shown in Fig. 1, three in the lower jawbones and one in the forehead. Once the simulation scan was finished, patient CT images were transferred to the Eclipse TPS (version 15.6, Varian Medical System Inc., USA) for structure contouring and treatment planning. Note that to provide the DICOM reference surface to AlignRT for surface tracking, the patient skinline named as BODY in TPS was first automatically extracted using an empirical predefined threshold of − 350 HU, and then carefully reviewed by physicist in case of any artifacts. Once approved by physicist and physician, treatment plans were transferred to linacs and AlignRT. At our institution, SRT treatments were all delivered on linacs using MLC-modulated RadipArc (Varian Medical System Inc., USA) with coplanar irradiation.
Treatment
AlignRT surface guidance system
The AlignRT surface guidance system at our institution was integrated with a VitalBeam (Varian Medical System Inc. USA) linac as shown in Fig. 2. The AlignRT system used three ceiling-mounted pods, each with a structured-light projector and two stereovision cameras. During surface imaging, projectors cast a speckle pattern onto the patient, and then the light pattern was captured by stereovision cameras to reconstruct the 3D real-time patient surface. The AlignRT software used a proprietary iterative closest points algorithm to register the real-time surface to the reference surface to derive relative displacement [10]. The derived surface displacement was displayed as six-degree-of-freedom (6DoF) real-time deltas (RTD) to indicate patient positioning errors.
Regular quality assurance is required for AlignRT performance maintenance. At our institution, daily QA, monthly QA, and isocenter calibration [24] as well as a self-developed weekly cube test [21] are periodically performed to ensure that the system works properly. The previous longitudinal study [21] shows that our AlignRT system is very stable and accurate to detect sub-millimeter and sub-degree shifts.
Patient setup and verification
The standard operation procedure at our institution for surface guided SRT patient setup is in shown in Fig. 3. Patient-specific DICOM-RT files were first imported into AlignRT prior to treatment, and the patient record, plan tree, reference surface and ROI were created in Record mode. In this study, the BODY structure extracted from the planning CT was used as the reference surface for all fractions. The ROI for surface tracking was drawn to include the opening facial area (the nose, eyes, and cheekbones) and exclude any part of the thermoplastic mask.
On treatment days, the specific patient in AlignRT was retrieved and the Treatment mode was activated to guide patient setup. First, the mask support with the posterior shell was indexed onto the setup board, and the patient was asked to lie down and therapists adjusted his or her posture to fit the headrest. Second, the anterior shell was carefully placed and secured with screws to minimize unbalanced pressure. The alignment of localization marks to room lasers was confirmed in the same process. Third, therapists activated the real-time monitoring function in AlignRT and carefully tuned the patient position in reference to the displayed RTD till the RTD values were within the institutional tolerance, i.e., 1 mm in translation and 1 degree in rotation. Next, a CBCT scan and CBCT-to-planCT registration was performed to verify tumor localization and derive couch shifts for position correction. If the couch shift was within the institutional tolerance, i.e., 3 mm in any translation direction and 2 degree in any rotational direction, the shifts would be applied. Otherwise, patient repositioning was mandatory to perform. Once patient setup was confirmed, a new reference surface was captured and used for patient motion tracking in this fraction only. After all the above operations were finalized, treatment delivery was initiated, and real-time patient motion was monitored by AlignRT with beam-hold control.
Recorded data and statistical analysis
Due to data missing in some cases, treatment records of 48 patients for 193 fractions in total were collected for this study, including: (a) RTD values in AlignRT at the end of patient initial setup; (b) setup corrections represented by CBCT couch shifts; (c) numbers of patient repositioning; (d) initial setup time, i.e., the time from patient lying down to therapists exiting the treatment room.
The AlignRT RTD values and CBCT setup corrections were first checked for normal distribution by Shapiro–Wilk test, and the results indicated normality was rejected (p value < 0.01) in several directions (CBCT-VRT, AlignRT-LNG, AlignRT-LAT, AlignRT-ROLL). Therefore, medians and percentiles of AlignRT RTD values and CBCT setup corrections were reported rather than means and standard deviations, and the difference in each direction was tested with the non-parametric Wilcoxon matched-pairs signed rank test. In this study, data analysis was performed in OriginPro (version 2021a, OriginLab, USA), and p value < 0.01 was considered statistically significant.