After receiving institutional review board approval, a retrospective study was conducted on 38 breast cancer patients treated with the Halcyon LINAC in our hospital. The treatment simulation was performed with a Brilliance CT Big Bore (Philips, Cleveland, OH, USA) using a slice thickness of 2.5 mm and a breast board was used for immobilization and to achieve reproducibility of the patient’s position. The Halcyon LINAC, which contains a 6 MV flattening filter free (FFF) and a dual-layer multi-leaf collimator (MLC), allows fast and accurate IMRT and VMAT based on a kV cone beam CT imaging (kV CBCT) system. Because it is necessary to move from the virtual isocenter to the treatment isocenter after the initial setup, 2D or 3D images should be acquired for every fraction to ensure that the positioning is accurate. Compared with the C-type LINAC, the patient setup for Halcyon treatment can be corrected with only translational error because the treatment couch has 3-DOF. The daily initial setup for each patient was performed by aligning skin markings with room laser for three fractions and by using SGRT for three fractions. In the LAS method, after marking the skin with ink at the reference isocenter during CT simulation, the patients were aligned to the marked locations and were manually shifted by the therapists. For the SGRT method, body surfaces were reconstructed using an imaging system that was used to match the body contours acquired from the reference CT image. After moving into the treatment isocenter, a total of 228 kV CBCTs scans were obtained for every fraction.
Sgrt Implementation
In our study, the AlignRT image-guided system (Vision RT Ltd., London, UK) was used for SGRT implementation. As shown in Fig. 1, AlignRT, consisting of a three-camera pod system, was installed on the ceiling and aligned to the virtual isocenter to monitor the patient’s initial setup out of the bore. System calibration was performed on the virtual isocenter owing to the area blocked by the O-ring type design. The daily QA for the SGRT system was performed using a plate with a circular pattern before treatment. When the maximum root mean square values representing the average discrepancy of the camera positions between the daily QA and system calibration exceeded 0.8 mm, the system was calibrated. While the initial setup of the patient was performed at the virtual isocenter, the reference body contour generated by the reference CT had the coordinates of the treatment isocenter; therefore, a dummy plan with a couch position without delta couch shift was generated using the Eclipse Scripting API. The region of interest (ROI) for breast cancer patients, which includes breast tissues, ribs, and the lateral aspect of the left and right breast to the midcoronal plane, was generated to monitor the difference in body contour quantitatively. The thresholds for the translational error (vertical, longitudinal, and lateral) and rotational error (pitch, roll, and yaw) were set to 3 mm and 3°, respectively. If the difference in the 6-DOF was within the thresholds, the couch was moved to the treatment isocenter for kV CBCT imaging.
Accuracy Verification For Rotational Error
The accuracy of the residual rotational error of the SGRT system was evaluated by using an in-house breast phantom. An anthropomorphic phantom (Model 702 Adult ATOM Female, CIRS Inc., Norfolk, VA), which is sectional in design with traditional 25 mm thick sections, was scanned with a 1.5 mm slice thickness. A phantom designed with a thickness of 25 mm has dosimetric capabilities, but it is difficult to obtain reproducible surface imaging. Therefore, to be suitable for the SGRT system, a phantom with an opaque/matte, light-colored surface that reflects the projected light pattern was fabricated using 3D printing. The body contours were converted into the standard tessellation language format for 3D printing. The phantom was manufactured from amorphous thermoplastic material using a Zortrax M300 3D printer (Zortrax, Olsztyn, Poland). The phantom on the baseplate was set to a virtual isocenter, and rotational errors from − 1.5° to 1.5° in steps of 0.5 °were induced for pitch, yaw, and roll using the baseplate, as shown in Fig. 2. The induced rotational errors were acquired using an SGRT system. Then, the rotational error was estimated by matching the body contour imported from Eclipse and acquired surface images using AlighRT, and the kV CBCT scans were acquired and imported into Velocity™ (Varian) to verify the residual rotational error using auto rigid registration between the kV CBCT and reference CT.
Data Acquisition & Analysis
Thirty-eight patients were investigated for a total of 228 fractions, including three fractions with the LAS method and three fractions with the SGRT method. The accuracy of the initial setup in terms of translational error was quantified by retrospectively analyzing the couch shifts between the reference CT and kV CBCT images. Because the Halcyon couch only moves the translational shift, both the kV CBCT and reference CT were exported into Velocity™ to compare the residual rotational error. The ROI was set to the appropriate dimensions in all three planes (axial, sagittal, and coronal) to include the chest wall and ribs to be fused. After fusing the two images, the rotational errors for the pitch, roll, and yaw were obtained from the automatic registration results. The setup efficiency was analyzed by comparing the initial setup and imaging times. The initial setup times were acquired by measuring the timestamp of the R&V system, starting from the instant the patient was loaded into the system and ending when the images were acquired. The imaging time was defined as from when the treatment beam was initialized after auto-and manual image registration was evaluated, and the necessary couch shift was performed. Patient data were analyzed by calculating mean, median, maximum, and quartile distributions. A paired t-test was used to establish the statistical significance of the results at P < 0.05.