Establishing an Offline Evaluation Process for IMRT/VMAT Radiation Therapy Treatment Planning Quality Using Knowledge-Based Planning: A Study Based on Multi-institutional Head and Neck Clinical Trial

Purpose/Objective(s): To build knowledge based planning model (KBP) for NRG-HN001 clinical trial, to establish offline quality assurance (QA) process for radiotherapy treatment planning with the aim of quality consistency, and to establish a quality threshold of sending the QA result back to institutions. Purpose/Objective(s): Fifty treatment plans of patients enrolled and treated using NRG-HN001 head and neck clinical trial which passed the contouring and dosimetric quality assurance criteria of the trial were used to build the KBP model in Varian Eclipse treatment planning system (TPS). The model was trained; verified using the standard procedure recommended by the vendor and was later used as a quality assurance tool for the treatment plans submitted to this clinical trial. Results: Thirty one treatment plans, which were not included in the training model, were re-optimized using the KBP model. The mean and median values for dose constraints of organ at risks (OARs) were lower in at least 23 of the re-optimized treatment plans as compared with those submitted by treating institutions, while maintaining good dose coverage for the PTVs. The mean and median maximum doses (in Gy) in the re-optimized and original plans for the selected critical organs were: 29.2/34.6, 19.94/36.16 of OpticNerve-R,; 29.87/35.8, 25.53/38.47 of OpticNerve-L, 27.3/32.6, 16.43/30.8 of OpticChiasm, 49.3/52.1, 49.58/51.1 of BrainStem, 40.3/42.1, 39.57/42.26 of SpinalCord, 70.63/70.84, 71.16/72.47 of Mandible, respectively. The average and median mean dose for the parotid glands were: Parotid-L (mean: 32.9/36.8, median: 29.71/32.84) and Parotid-R (mean: 29.9/31.3, median: 28.08/31.78) in the re-optimized and original plans, respectively. The average values for the maximum dose percentage differences in the re-optimized and original plans were 19%, 19.5%, 21%, 5% and 4% for OpticNerve-R, OpticNerve-L, OpticChiasm, BrainStem and SpinalCord respectively and these differences were statistically significant (P< 0.05). Conclusion: A KBP model was built in Eclipse TPS and used as offline quality assurance tool to evaluate the quality of treatment plans submitted to the NRG-HN001 clinical trial. Dose parameters for OARs in at least 23 treatment plans out of the 31 re-optimized plans were improved. The model will be used continuously as a QA tool of future treatment plans of patients accrued in the trial. Submitting institutions would be notified with any improvements of their plans of ≥ 5% in one of the OARs that are given high priority in planning optimization (optic organs, brain stem and spinal cord) provided that other target and OAR dosimetry are approximately equivalent.