Purpose/Objective(s):
MRI-guided online adaptive radiotherapy enables online re-optimization based on daily anatomical changes. However, significant variations in online planning and delivery times have been observed, suggesting a lack of robustness and inefficiency in the online process. This study tests the hypothesis that reference plan parameters are a primary source. By retrospectively analyzing inter-plan variations and their impact on planning time and delivery efficiency, we identified key factors contributing to online instability. Furthermore, we introduced an anatomy-informed plan parameter selection method that leverages high-quality historical data to recommend optimal parameters, ultimately enhancing online workflow robustness and efficiency.
Materials/Methods:
Thirty-three pancreas (SIB 5000/3300cGy or 4000/2500cGy) patients treated with MRI-Linac were retrospectively analyzed (five fractions per patient). Online planning times were recorded using our in-house ART dashboard, and delivery time was estimated from total segments and MUs using a validated fitting equation (=1-minute error), excluding gating duty cycle effects. For each patient, the mean and standard deviation (SD) across five fractions were calculated; higher means indicated lower efficiency, and larger SDs reflected greater variability and reduced robustness. Seven plan parameters were extracted and correlated with online performance variation. Anatomical complexity was quantified by PTVs volume and OAR (duodenum, stomach, small bowel) overlap with the PTVs. Patients were ranked within each prescription group using a composite score (planning time + estimated delivery time + inter-fraction SD), scaled by PTV volume. The three lowest-ranked cases were matched to top-ranked cases based on anatomical similarity (PTV and OAR overlap). High-performing plan templates were applied to generate new plans, and adaptive plans were created on three daily images. Planning time was recorded, delivery time was calculated, and overall performance was evaluated.
Results:
Significant inter-patient variability was observed, with segments ranging from 69–189 and total MUs from 1629–6888. Mean planning time per patient was 9.8 ± 4.6 mins (5.6–15.1), and mean delivery time was 20.0 ± 1.2 mins (13.0–25.0). Wider segment widths (0.7–1.0 cm) shortened delivery time but reduced robustness, increasing the planning time SD. The high-performing template based new plans maintained comparable plan quality. For the adaptive plans, planning plus delivery time was reduced by 9.1 ± 2.9 mins, with improved stability (SD=1.3 ± 0.2 vs 13.3 ± 1.1 before).
Conclusion:
Our results demonstrate that plan parameters substantially impact online adaptive workflow performance. Anatomy-informed, knowledge-based optimal plan templates provide a practical strategy to enhance consistency, online robustness, and efficiency. Future work will focus on developing more advanced, intelligent anatomy-aware planning frameworks.
