ESTRO 2024 - Abstract Book

S4538

Physics - Machine learning models and clinical applications

ESTRO 2024

Conclusion:

The models trained work relatively well for TB0 and TB1 and, therefore, a model trained in one can predict pre treatment verification outcomes in the other. As could be expected a model trained in a TrueBeam does not work properly in a Clinac. However, we didn’t expect the results for TB3. Further analysis showed that the underlaying complexity metrics for TB3 behaved differently than for TB0 and TB1.

In conclusion, we recommend to validate model transfer between LINACs, even of the same model, before using a classifier in routine.