ESTRO 2024 - Abstract Book

S3090

Physics - Autosegmentation

ESTRO 2024

Conclusion:

We have constructed a 3D Unet for CTV segmentation trained on 611 patients of a single institute. The training performance was accessed through cross-validation metrics, obtaining DSC: 0.903 and HD_95: 9.1 mm. This result was confirmed by the average metrics obtained in the internal test. Future work will compare current results against different architectures (nnUnet, ResNet) and optimize their hyperparameters with Optuna.

Keywords: breast cancer, deep learning, clinical use

References:

This study is supported by ERAPERMED-2020-110-JTC.

1. Men K. et al. Physica Medica 50 (June 2018), p. 13–19. doi: 10.1016/j.ejmp.2018.05.006

2. Eldesoky A. R. et al. Radiotherapy and Oncology 121.3 (Dec. 2016), p. 424–430. doi: 10.1016/j.radonc.2016.09.005