e23158 Background: Inter-fractional variability of pelvic organ anatomy introduces non-linear geometric uncertainty that challenges the reliability of conventional margin-based radiotherapy planning. Online adaptive radiotherapy (oART) enables real-time plan modification to address such variability. This study evaluates the operational robustness and geometric dose-stabilization capability of an artificial intelligence–driven oART workflow under routine clinical care conditions. Methods: A retrospective fraction-level analysis was performed on 185 treatment fractions from 9 patients with prostate cancer treated using moderately hypofractionated oART (60–68 Gy). Rectal and bladder anatomical variation was quantified for each fraction using absolute volumetric deviation (cc) relative to reference planning geometry. Fractions were categorized by deformation severity as minor ( 30 cc). Workflow performance and dose-control reliability were evaluated using D15%, defined as the dose received by 15% of the organ-at-risk (OAR) volume, extracted from dose–volume histograms for each fraction. Fraction-level dosimetric impact was assessed by comparing adapted plans with corresponding non-adaptive scheduled plans (ΔD15% = Adapted − Scheduled). Results: The oART workflow demonstrated 100% mechanical and delivery reliability across all 185 fractions. Moderate-to-major anatomical deformation (≥10 cc) was observed in 118 fractions (63.8%), including 53 fractions (28.6%) with major deformation ( > 30 cc). Mean rectal and bladder volume changes were 12.8 cc and approximately 75 cc, respectively. Despite substantial anatomical variability, oART stabilized high-dose organ exposure. Median ΔD15% was −1.0 cGy for rectum and approximately +5 cGy for bladder across all fractions. Increasing deformation magnitude did not result in proportional dose escalation, indicating effective truncation of extreme organ-at-risk dose excursions that would otherwise reach up to ~70 cGy per fraction in non-adaptive scenarios. Conclusions: In the presence of frequent and clinically significant inter-fractional pelvic anatomical deformation, artificial intelligence–driven online adaptive radiotherapy demonstrated robust workflow reliability and effective geometric dose stabilization. These findings support oART as a reliable care-delivery strategy providing non-linear geometric dose control beyond conventional margin-based planning in moderately hypofractionated prostate radiotherapy.
Shaikh et al. (Thu,) studied this question.