ESTRO 2021 Abstract Book

S523

ESTRO 2021

Conclusion A relatively small training dataset is sufficient to train DLAS in a clinical setting. This is an important finding, as it greatly simplifies clinical implementation of DLAS using institute-specific DL models. Training datasets larger than 20 patients hardly improved the segmentation results, even for structures with moderate CT contrast. Furthermore, deep learning based auto segmentation outperformed atlas based auto-segmentation in breast cancer patients, when using training data set of ≥ 20 patients. PH-0654 End-to-end head & neck tumor auto-segmentation using CT/PET and MRI without deformable registration J. Ren 1 , J.A. Nijkamp 2 , J.G. Eriksen 3 , S.S. Korreman 1 1 Aarhus University, Department of Clinical Medicine - The Department of Oncology, DCPT - Danish Center for Particle Therapy, Aarhus, Denmark; 2 Aarhus University, Department of Clinical Medicine - DCPT - Danish Center for Particle Therapy, Aarhus, Denmark; 3 Aarhus University, Department of Clinical Medicine - Department of Experimental Clinical Oncology, Aarhus, Denmark Purpose or Objective Deep learning-based tumor segmentation is expected to alleviate time consumption and inter-observer variability (IOV) of manual delineation by learning complementary information from multimodality images. Currently, multimodality images are required to be co-registered prior to segmentation by convolutional