Population-Based Evaluation of Rectum-Sparing Margins In Prostate SBRT
Abstract
Purpose
Prostate SBRT demands high geometric precision due to large fractional doses and steep dose gradients. Conventional PTV margins are typically derived from population-level estimates and may not reflect individual patient-specific motion. This work introduces a Bayesian hierarchical framework to simulate intrafraction motion and evaluate its dosimetric consequences.
Methods
A cohort of 205 patients treated with five-fraction prostate SBRT from 2020 to 2024 were retrospectively analyzed. Mid-treatment shifts were extracted from CBCT records and used to represent unintentional motion during treatment. These data trained a multivariate Student-T model with partial pooling, capturing population-level trends while retaining individual variability. Posterior predictive sampling generated synthetic shifts, which were applied to patient dose grids using rigid transformations. Population-based cumulative dose distributions were recalculated and evaluated using standard DVH metrics.
Results
Simulated shifts reproduced the central tendency, spread, and directional correlations of recorded patient data. Failure rates under clinical margins closely matched those in patient records, and correlation structures between axes were preserved. Population-level analysis showed very close agreement in DVH deviations across prostate, bladder, and rectum when simulated for 5 fraction deliveries.
Conclusion
This modeling framework offers a flexible approach to reconstructing patient-specific motion, even with limited empirical data. By leveraging partial pooling, it balances individual detail with population-informed inference. Its ability to replicate real-world distributions, failure rates, and dose perturbations underscores its validity as a simulation engine. Importantly, this approach enables probabilistic “what-if” testing without requiring additional imaging or patient intervention—offering a principled method for stress-testing clinical practices. In doing so, it advances clinical conversation around personalization in radiation therapy and highlights the role of statistical modeling in improving treatment, safety, and precision.