Evaluation of Rapidplan Model Performance after an Eclipse Version Upgrade
Abstract
Purpose
RapidPlan, a knowledge-based treatment planning tool integrated in Varian’s Eclipse treatment planning system, has been shown to improve radiotherapy planning efficiency and consistency. The aim of this study is to evaluate the performance of a RapidPlan model following an Eclipse version upgrade from v15 to v18.
Methods
The RapidPlan model was developed for five-fraction lung SBRT in Eclipse v15, using 2-3 VMAT arcs. This model was retrospectively applied to 15 lung cancer patients treated between 2023 and 2025. The prescriptions were 50 Gy, 55 Gy or 60 Gy in 5 fractions. Target volumes were 5.4-70.3 cc. The plan generated by the RapidPlan model in Eclipse v15 was compared to that generated in Eclipse v18. To facilitate comparison, same VMAT arc geometries and model parameters were employed. No manual intervention to the optimization other than normalizing prescription dose to cover 95% of the target. Following plan quality metrics were analyzed: applicable clinical objectives (as defined in RTOG0813 and AAPM TG101), total MU, maximum dose (Dmax), maximum dose 2cm away from the target (D2cm), lungs V20Gy, conformality index and gradient measure as reported in Eclipse, using Wilcoxon signed-rank test.
Results
For applicable clinical objectives, the plans generated in Eclipse v18 demonstrated the same pass rates as those generated in Eclipse v15 across all cases. No statistically significant differences in the medians of the remaining metrics were observed at the 1% significance level, except for Dmax (Median v15: 128.6%; v18: 124.7%).
Conclusion
After Eclipse v18 upgrade, the five-fraction lung SBRT RapidPlan model produced plans of comparable quality to those generated in Eclipse v15. The difference in Dmax is not clinically meaningful. These results suggest that RapidPlan models developed in a previous Eclipse version can be effectively reused to provide a viable starting point for plan optimization after an Eclipse upgrade.