Room: Stars at Night Ballroom 2-3
Purpose: To determine the prediction accuracy of a model-based selection tool in identifying Locally Advanced Pancreatic Cancer (LAPC) patients benefiting from plan adaptation while being treated with SBRT.
Methods: A total of 133 high-resolution CT scans, consisting of the planning CT (pCT) and ideally 3 daily CTs from 35 LAPC patients, were used as basis for analyzing daily variations observed in the organs-at-risk (OAR) (duodenum, stomach and bowel). For each patient, a population-based statistical model, based on principal component analysis and trained with the daily variations of the remaining patients, was used to simulate potential OAR deformations of the pCT anatomy. Dose-volume histograms obtained from the original treatment plan sampled on the simulated organs resulted in the probability of exceeding the dose-constraints due to anatomical variations. A threshold was established per organ to maximally cluster the simulated probabilities according to clinical observations. If the prediction of the three OAR for a patient was above the three established thresholds, the patient was considered requiring plan adaptation, while otherwise not. A geometric discrimination as suggested in the literature was tested against the model-based prediction to assess the added value of our approach.
Results: Clinical observations revealed that in 21/35 patients at least one OAR exceeded the dose-constraints in the daily CTs. Our model-based prediction compared to the geometric assessment identified 4 more patients not benefiting from plan adaptation; resulting in a sensitivity and specificity of 100 and 64%, and 100 and 36%, and therefore, in overall prediction accuracies of 86 vs. 74%, respectively.
Conclusion: A model-based prediction tool learning from OAR variations results in a more robust approach identifying patients benefiting from plan adaptation than using geometric metrics, which only evaluate the patient anatomical configuration. Therefore, our model-based approach might be a potential tool to optimize the clinical resources for LAPC patients.
Funding Support, Disclosures, and Conflict of Interest: This work was in part funded by a research grant of Accuray Inc., Sunnyvale, USA. Erasmus MC Cancer Institute also has a research collaboration with Elekta AB, Stockholm, Sweden.