Constructing and Validating a Knowledge‑Based Planning Model for Lung SBRT In Eclipse
Abstract
Purpose
To develop and clinically validate a RapidPlan knowledge-based planning (KBP) model for lung Stereotactic Body Radiotherapy (SBRT), aiming to improve plan quality, consistency, and efficiency in routine clinical practice.
Methods
A RapidPlan model was trained on 161 coplanar volumetric modulated arc therapy (VMAT) SBRT plans calculated with Acuros XB (2 mm grid; 6 MV FFF), covering six prescription regimens from 30 Gy ×1 to 60 Gy ×8. Model Analytics assessed dose-volume histogram (DVH) prediction bands, sigma index homogeneity, and geometric/dosimetric outliers; four atypical plans were excluded. An optimization template incorporated explicit target priorities, a planning target volume (PTV) upper generalized equivalent uniform dose (gEUD, a=20) with lower priority, an internal target volume (ITV) lower objective with higher priority, and gradient-oriented constraints for Dose at 2 cm from the PTV (D2cm) and ribs, along with a manual Normal Tissue Objective (NTO, Priority 145, Start 100%, End 40%, Fall-off 0.20). Five representative cases were used for iterative tuning. Prospective validation on 27 patients compared KBP plans with previously approved clinical plans, recording first-pass acceptability and required adjustments.
Results
Prediction bands were stable for ITV (mean 114%, SD = 1.12) and PTV (mean 108%, SD = 2.25). 62.5% of KBP plans met clinical criteria on first optimization, and 81% were equal to or better than prior clinical plans. Remaining deviations, mainly brachial plexus or spinal canal Dmax or PTV undercoverage, were corrected with minor edits in <10 minutes. The combination of lower-priority PTV gEUD and higher-priority ITV lower objective consistently maintained global maximum dose within ITV in 24 of 27 plans, without cropping the ITV to the PTV.
Conclusion
The model provides a reproducible, clinically validated framework for lung SBRT KBP, supporting consistent plan quality, efficient workflow, and reduced manual effort, with robust gradient and hotspot control across diverse prescri ptions.