Room: ePoster Forums
Purpose: The purpose of this work was to determine if separate knowledge-based models based on location of the secondary targets were necessary to generate clinical quality plans.
Methods: Using RapidPlan (RP), two models, ipsilateral and bilateral, were created for head and neck cancer based on the laterality of the secondary nodal disease. Varian provided analysis tools were used to identify outliers, which were subsequently removed. Model validation consisted of DVH comparison to clinical plans. This process utilized plans excluded from the model that were created by a single iteration of the optimizer. Validation metrics were derived from clinical planning protocol objectives, which also served as the basis of the optimization objectives for inverse planning. These metrics included the maximum dose, D99% and D15% of the PTV High volume, and D95% for the PTV Intermediate and PTV Low volumes. Fifteen dose metrics for organs-at-risk (OAR) were also evaluated. Finally, the performance of both models was compared for an ipsilateral nodal disease subset.
Results: Each model generated plans that typically met most of the objectives, but systematically delivered higher maximum doses than desired. A maximum dose exceeding 115% of prescription dose was observed in 25% and 70% of the plans created with the bilateral and ipsilateral RP models, respectively. Compared to the non-RP clinical plan, dose to 11 out of 15 OARs was lower using the bilateral RP model and 12 out of 15 using the ipsilateral model. When the bilateral model was applied to ipsilateral cases, the average difference between them over all objectives was 0.11% +/- 4.3 (1 SD).
Conclusion: The bilateral model handled ipsilateral nodal cases as well as the model specifically trained for those cases. Furthermore, on average, the dose to the OARs could be reduced over the non-RP clinical plans by applying the bilateral RP model.