Screen: 9
Eric Paulson, PhD
Medical College of Wisconsin
Milwaukee, WI, United States
Materials/
Methods: An anthropomorphic phantom and three heathy volunteers were included in the study. Abdominothoracic breath held 3D T1 Dixon images were acquired at end expiration on a 1.5T MR Linac. Target volumes (vertebral bodies) were constructed on in-phase Dixon images. Deep learning (DL)-based synthetic CT images were derived from standardized, opposed-phase Dixon images using a 2D cycleGAN model trained using 154 images. The synthetic CT was then thresholded to create air, bone, and unspecified tissue contours, which were then transferred to the in-phase Dixon image. Reference plans designed to deliver 800 cGy in one fraction were created on the in-phase Dixon images using a 3D AP/PA beam arrangement in offline Monaco, with electron densities (ED) extracted from the synthetic CT. Approved reference plans were then promoted and configured in Mosaiq. A treatment session was initiated and in-phase 3D T1 Dixon verification images were acquired and sent to online Monaco. The Adapt-To-Position (ATP) workflow was performed using segment aperture morphing (Adapt Segments). Approved adaptive plans were then downloaded and delivered. The total time required to perform the workflow was recorded.
Results: The total time to complete the online MRgRT scan-plan-treat workflow, including reference and verification imaging, synthetic CT generation, contouring, reference and adaptive planning, software-based QA checks, and delivery of 800 cGy, was 18.5 minutes in phantom and 23.5 +/- 2.9 minutes (mean +/- standard deviation) for the healthy volunteers. Use of the DL synthetic CT removed dependence on prior diagnostic CT images for ED assignment. Average reference and ATP plan Dmax values were 113.2% and 114.2% of the prescription dose, respectively. Average reference and ATP plan target D95% values were 755.4 cGy and 759.1 cGy, respectively. Small shifts (<5 mm), introduced through the automated couch insertion process in offline Monaco and detected when registering verification images to reference images in online Monaco, were compensated with the Adapt Segments mode.
Conclusion: Emergency palliation using an online MR-guided scan-plan-treat workflow is feasible. Despite increased complexity, online MRgRT scan-plan-treat workflows can be performed in under 30 minutes, minimizing resources, time, and number of table transfers for patients in pain.