Program Information
Investigating the Effect of Dosimetric Outliers On the Performance of RapidPlan Models: A Study Based On Multi-Institutional NRG-HN001 Head and Neck Clinical Trial
T Giaddui1*, H Geng1 , Q Chen2 , M Radden3 , N Linnemann3 , N Lee4 , P Xia5 , Y Xiao1 , (1) University of Pennsylvania, Philadelphia, PA, (2) Geisinger Commonwealth School of Medicine, Scranton, PA, (3) IROC- Philadelphia- RTQA Center, Philadelphia, PA, (4) Memorial Sloan Kettering Cancer Center, New York, NY, (5) The Cleveland Clinic Foundation, Cleveland, OH.
Presentations
SU-I-GPD-T-442 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: To investigate the effect of removing dosimetric outliers on the performance of RapidPlan models.
Methods: Fifty treatment plans of patients enrolled and treated using NRG-HN001 head and neck clinical trial were used to build an NRG-HN001 specific RapidPlan model in Varian Eclipse treatment planning system (TPS). Dosimetric outliers were initially identified from the model statistics summary. Plans with dosimetric outliers were then re optimized using the model. We duplicated the model, re added the improved re optimized plans and deleted dosimetric outliers (outlier structures) from the model when no improvement was achieved from the re optimization and saved as a new model. The cleaning processes went through three steps and three new models were trained. We compared the performance of the three new models with the original model through re-optimizing verification treatment plans (other than those used to build the model).
Results: Dose volume histograms of five structures (brain stem, spinal cord, optic chiasm, optic nerves, parotid glands) in the re-optimized plans were compared using the four models (original model and the three trimmed models) to assess the effect of the trimming process on the performance of the RapidPlan models. The differences in the maximum dose of different structures of the three trimmed model as compared with original model varied between: - 6.66 Gy to 11. 3 Gy for brain stem; -11.38 Gy to 8 Gy for spinal cord; -13.52 Gy to 0.22 Gy for optic chiasm; -18.5 Gy to 2.27 for Optic nerves. The differences in the mean dose of the parotid glands varied between – 4.88 Gy to 4.7 Gy
Conclusion: Model performance can be affected by removing outliers and these effects can be significant for some treatment plans. Removing outliers does not always improve the performance of the RapidPlan models for all structures.
Funding Support, Disclosures, and Conflict of Interest: This project was supported by grants U10CA180868 (NRG Oncology Operations), U10CA180822 (NRG Oncology SDMC), U24CA180803 (IROC) from the National Cancer Institute (NCI).
Contact Email: