ESTRO 2024 - Abstract Book

S3580

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2024

Main results of the cross validation tests are shown in the Figures. Average values of the prediction bands on the test patients (n=30/25 for right/left breast respectively) showed large overlap with the prediction bands of each institutional models: for right breast, KB-Benchmark model predicts slightly lower mean dose to ipsilateral lung when compared to every KB-institutes model, excepting for INST7 (Figure 1).

Regarding the left breast, even better similarity between KB-Benchmark and KB-institute models’ predictions were found for ipsilateral lung.

Concerning heart DVHs, large overlaps were similarly found between KB-Benchmark and KB-Institution predictions (Figure 2): the spread of KB-Benchmark predictions were larger compared to band predictions of KB institute models in the low dose range. This result was interpreted as likely due to INST9 whose heart volumes were significantly smaller than the other Institutes.

Conclusion:

Results of KB-institutes models analysis showed a limited inter-institute variability of plan prediction models and a satisfactory inter-institute transferability, excepting one institute for right and one for left breast. These achievements have paved the way for generating two KB-Benchmark models, whose performances sufficiently match those of each KB-institute model. These KB-benchmark models could be used for plan prediction in WBI with potentially relevant applications for large-scale KB-driven automatic plan optimization, QA audit and tutoring/education purposes.