Presenter(s)
E. A. Omari, P. W. Prior Jr, X. Chen, and E. S. Paulson; Department of Radiation Oncology, Medical College of Wisconsin, Milwaukee, WI
Purpose/Objective(s): Palliative simulation free radiation therapy (SFRT) has gained interest in recent years. SFRT can play an impactful role in reducing CT simulation load, wait times, resource demands, and number of table transfers for patients in pain. Diagnostic CT has been utilized clinically for SFRT dose calculation. However, differences in CT calibration, patient positioning, iodinated contrast, and scanning protocols (field of view, slice coverage, gantry tilt) compared with CT simulators can add complexity and may introduce dosimetric uncertainties. In this work, we investigate the feasibility to utilize onboard kV CBCT images for online SFRT.
Materials/Methods: An online SFRT workflow was developed and tested on a C-arm linac (Elekta Versa HD) equipped with onboard kV CBCT. The workflow consist of 0) pre-populating patient information in Mosaiq, 1) patient setup, 2) CBCT acquisition, 3) deep learning (DL)-based synthetic CT (sCT) image generation in ADMIRE (Elekta), 4) DL-based auto-segmentation (DLAS) in MIM (MIM Software), 5) target delineation, 6) plan generation and review using density overrides in Monaco (Elekta), 7) secondary dose calculation (RadCalc), 8) plan promotion in Mosaiq, 9) software-based QA, 10) portal imaging, and 11) treatment delivery. SFRT plans were designed to deliver 800 cGy in one fraction using 3D techniques. A retrospective analysis of DL-sCT images generated from CBCT images acquired from 40 thoracic and pelvis cancer patients was performed (IRB: PRO00024935). The DL-sCT was introduced to achieve higher DLAS accuracy compared to clinical CBCT images. However, SFRT plans were generated on CBCT images with density overrides, analogous to our established MR-linac workflows. End-to-end testing was performed using an anthropomorphic phantom. Total time required to perform the SFRT workflow was recorded.
Results: sCT images generated by the DL models (<20 seconds) exhibited image quality and CT number accuracy comparable to simulation CT images, meeting published tolerances for < 1% dose deviation. This facilitated rapid DLAS (<30 seconds) and precise dose calculation, with a 3D gamma passing rate of 97±2% (1%/2mm criteria). sCT demonstrated similar DLAS accuracy to the planning CT, with dice similarity coefficient > 0.8 and mean distance to agreement < 2 mm for most structures. The designed end-to-end workflow averaged at 30 minutes. Efficiencies were introduced by using DLAS models and generating templates for plan creation.
Conclusion: Online SFRT with onboard kV CBCT images is feasible, without the need for pre-planning or use of diagnostic images. The proposed workflow enables a significant time reduction compared to traditional treatment planning approaches; minimizing resources, and number of table transfers for patients in pain. Future work will focus on clinical translation and the development of auto-planning capabilities, with the goal of further expediting the workflow and broadening its utility to include curative-intent treatments.