Session Item

The study aimed to evaluate the results of knowledge-based planning models clinically validated in 6 sites on a non-homogeneous group of gynaecological patients as part of the model quality control (QC) program.

Six centres created and clinically validated RapidPlan (VARIAN) models for their own group of gynecologic patients. Each site then selected ten randomised gynaecological patients, other than those used in the model, anonymised and exported them. Patients were delineated according to their institution’s internal protocols: CTV, PTV, rectum, bladder, left and right femoral heads (FH), bones and bowels. Each site then prepared plans for a group of all 60 patients following its procedures. The Eclipse TPS was used (AAA - 4 sites, Acuros XB - 2 sites). The VMAT technique, X 6MV photon beams and TrueBeam machine (Varian) with MLC120 were used. Based on the model predictions, one optimisation was performed without the planner’s intervention. Data from 6 treatment plans were collected and compared for each patient. For CTV and PTV: D98, D95, D2, Dmax, V95% were analysed. For OARs, the following statistics were compared: rectum - V10, V20, V30, V40, V50, D0.03cm³, bladder - V20, V45, V50, D0.03cm³, bowels - V45, FH L/R - V35, Dmax, bones - Dmean. The statistical significance of the differences was tested with the Kruskal-Wallis test (p<0.5).

For CTV only in one model (S5), parameter D98% > 95% was violated in 4 cases. In the rest, it was always fulfilled. For PTV, the lowest coverage was achieved by the S2 model (on average D98% = 94.3%). For the S1, S3, S4, S5 and S6 models, the average was 95.7%, 97.3%, 95.6%, 95.5%, and 98.8%, respectively. For the rectum and clinically meaningful parameters D0.03cm³ and V50Gy, the maximum difference between models was on average 0.8 cm³ and 1.45 Gy, respectively. Larger differences occurred in the moderated doses, reaching on average 17.7% for V30. The same was true for the bladder, where the models’ results did not differ clinically in high doses. The maximum difference for D0.03cm³ was 0.4 Gy. In contrast, in moderated doses, the differences reached on average 26.3% (V20). For the bowels, the parameter V45 > 195 cm³ was most often exceeded in S3 (22 times) and the least frequently in S2 (11 times). All models failed to meet the bowels constraints for the same 11 patients (the anatomical reasons). For bones, the maximum difference in mean dose was on average 3.4 Gy. Dmean > 25 Gy less often was violated for S5 (24 times) and the most for S3 (55 times).

The differences obtained by the six models were statistically significant, but the greatest differences were found in parameters not clinically relevant (moderated doses). The results were not influenced by the contouring method but by the requirements for dose distribution adopted at the site and the quality of learning plans. It seems that the mutual validation of models is valuable for an arbitrary assessment of their quality.