Simulation-Free Adaptive Prostate SBRT on MR-Linac: Technical Feasibility and Clinical Experience.
Authors
Affiliations (3)
Affiliations (3)
- Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, USA; Université de Montréal, Department of Physics, Montréal, QC, Canada. Electronic address: [email protected].
- Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, USA.
- Université de Montréal, Department of Physics, Montréal, QC, Canada.
Abstract
To develop and evaluate two simulation-free (SF) stereotactic body radiotherapy (SBRT) workflows for prostate cancer on Magnetic Resonance (MR)-Linac: one using a diagnostic MR reference scan and another using a pre-approved template scan. Population-based relative electron density (RED) values were derived from 100 prostate SBRT patients on the Elekta Unity MR-Linac and validated on a separate cohort of 10 patients for dosimetric equivalence. Two SF approaches were evaluated in another independent cohort of 10 patients using reference plans derived from (1) a diagnostic MR scan (5 retrospective, 5 prospective) and (2) a digitally created pelvic phantom scan using population-based RED (all retrospective). The performance of Artificial Intelligence (AI)-generated contours was investigated during adaptive treatments using surface Dice (sDSC) with 1 mm tolerance and Added Path Length (APL). All adaptive plans, from both SF and the clinically delivered workflow, were compared in terms of dose metrics and monitor unit (MU) usage. A focused Failure Modes and Effects Analysis (FMEA) identified and mitigated diagnostic MR SF specific risks by calculating the Risk Priority Number (RPN). Population-based RED assignments resulted in mean dose differences under 1%. AI contours showed strong agreement with physician-approved contours, with sDSC and APL of 0.96/26.2cm,0.98/17.2cm,0.99/0.4cm, and 0.99/0.01cm for bladder, rectum, penile bulb, and bone respectively. Across both SF workflows, adaptive plans were clinically comparable to the standard workflow, with mean dose differences from -1.7% to 4.6% and mean MU differences remaining under 2%. FMEA identified three high-risk failure modes: mislabeling diagnostic MR scans (RPN=39), case tracking lapses (RPN=29), and incorrect RED assignment (RPN=22), all mitigated through workflow refinements and training. The diagnostic MR SF workflow has been clinically implemented, treating 22 patients without failures. Both SF workflows eliminated the need for simulation, streamlining adaptive prostate SBRT without compromising plan quality. The template-based workflow further improved patient access by removing imaging prerequisites.