Safety and Precision In a Same-Day, MRI-Only Simulation with Adaptive VMAT SRS/SRT Workflow: Integrating Synthetic CT and AI-Driven Quality Assurance
Abstract
Purpose
The efficacy of stereotactic radiosurgery (SRS) and radiotherapy (SRT) for brain metastases is often compromised by tumor growth and soft tissue changes between simulation and treatment. To eliminate these latencies, we clinically implemented a novel same-day workflow combining MRI-only simulation with non-coplanar volumetric modulated arc therapy (VMAT) for intracranial SRS/SRT. Here we report on the workflow design, commissioning, and the integration of a deep learning-based auto-segmentation tool as a Quality Assurance (QA) safety net.
Methods
On the day of treatment, patients undergo MRI simulation in the treatment position. Target definition is completed on a T1-weighted MPRAGE MRI. To mitigate the risk of missed lesions during the accelerated planning timeline, an AI auto-contouring prototype by Siemens was integrated to provide a "second check" for GTV detection and OAR delineation. Treatment planning is completed using MRI-defined targets with dose calculation and reference image generation based on a deep-learning-based FDA-approved synthetic CT (sCT) solution. A preliminary plan based on diagnostic imaging is adapted to the treatment-day anatomy and re-optimized (HyperArc/VMAT). Commissioning included: 1) dosimetric comparison of sCT vs. CT plans, 2) verification of CBCT-to-sCT positioning accuracy, and 3) validation of the AI-QA tool's sensitivity on a retrospective cohort of 436 patients.
Results
Dosimetric analysis demonstrated a D95% dose difference of -0.04±0.72% between sCT and CT plans, with Gamma indices exceeding 99% (1%/1mm). CBCT-to-sCT alignment achieved sub-millimeter accuracy. The AI-QA demonstrated a sensitivity of 93.3% for detecting metastases >0.1 cc with robust segmentation (median Dice >0.79).
Conclusion
The same-day MRI-only workflow is clinically feasible and robust. The integration of AI-driven contouring QA provides a safety layer, addressing the challenges of accelerated timelines by ensuring comprehensive target delineation. A trial to assess whether this workflow may allow for the safe omission of PTV margins through adaptation to anatomical changes is currently underway.