ESTRO 2024 - Abstract Book

S3005

Physics - Autosegmentation

ESTRO 2024

Conclusion:

The addition of a patient-specific prior in the form of a probability map makes a CTV segmentation model more robust, since it implicitly takes the clinical decision making into account. Performance of a model with prior trained on external data is similar to model performance without prior trained on internal data, showing that the model with patient-specific prior is suitable for generalization to multiple centers.

Keywords: Deep learning, online ART, generalizability

448

Mini-Oral

National consensus-based bespoke automatic delineation of thoracic organs at risk

Lise BJ Thorsen 1,2 , Hanna R Mortensen 3 , Lone Hoffmann 1,2 , Torben Aagaard 1 , Marianne M Knap 1 , Tine B Nyeng 1 , Weronika M Szejniuk 4,5 , Gitte Persson 6 , Mette Pøhl 7 , Marianne Nordsmark 1,2 , Britta Weber 3 , Charlotte Kristiansen 8 ,