A Simulation-Free Radiation Therapy Workflow Using Synthetic Computed Tomography Generated from Diagnostic Magnetic Resonance Imaging for Personalized Hippocampal-Sparing Whole-Brain Treatment.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiation Oncology, UT Southwestern Medical Center, Dallas, Texas; Medical Artificial Intelligence and Automation (MAIA) Lab, UT Southwestern Medical Center, Dallas, Texas.
- Department of Radiation Oncology, UT Southwestern Medical Center, Dallas, Texas.
- Department of Radiation Oncology, UT Southwestern Medical Center, Dallas, Texas; Medical Artificial Intelligence and Automation (MAIA) Lab, UT Southwestern Medical Center, Dallas, Texas. Electronic address: [email protected].
Abstract
The conventional computed tomography (CT)-based consultation to simulation process for hippocampal-sparing whole-brain radiation therapy (HS-WBRT) typically requires several days, delaying treatment initiation for patients who would benefit from expedited care. We developed a simulation-free, magnetic resonance imaging (MRI)-only workflow for HS-WBRT to accelerate treatment start and maintain dosimetric quality through online adaptive radiation therapy. An in-house deep learning algorithm was developed to generate synthetic CT (sCT) images from diagnostic MRI scans. To address the limited field-of-view of diagnostic MRI, a hybrid sCT-reference CT stitching method was developed to create a full planning sCT suitable for HS-WBRT, through preserving patient-specific brain anatomy while supplementing inferior regions with matched reference anatomy. Using the resulting sCT, a patient-specific reference plan was created and then adapted to the setup position and patient anatomy during the first fraction. Subsequent fractions were delivered using the first-fraction adapted plan or readapted as needed. Commissioning included 5 retrospective emulator cases and a phantom end-to-end test to validate workflow feasibility, plan quality, and interoperability. Clinical feasibility was prospectively assessed in 5 patients. All retrospective and phantom tests met predefined endpoints, including protocol compliance, secondary quality assurance ≥95% (3%/2 mm), and error-free data transfer. In the prospective cohort, hippocampal contours were finalized in all cases during the online session without resimulation. The first-fraction adaptive plan met all HS-WBRT protocol constraints. For patients prescribed 30 Gy in 10 fractions (n = 4), median hippocampus D<sub>max</sub> and D100% were 14.6 Gy (range, 13.0-15.7 Gy) and 7.2 Gy (range, 7.0-7.5 Gy), respectively. For the patient prescribed 20 Gy in 10 fractions, hippocampus D<sub>max</sub> and D100% were 14.6 Gy and 5.2 Gy. All other organs-at-risk met institutional limits. The workflow eliminated CT simulation, enabling a clinical approved plan within 2 days of diagnostic imaging. Median on-couch adaptation time was 39 minutes. This MRI-only, simulation-free workflow is clinically feasible, shortens time to treatment, and enhances patient experience without compromising dosimetric quality.