ESTRO 2021 Abstract Book

S270

ESTRO 2021

pCT to 0.79 for the shrinkage estimate. The improvement tended to be greater for patients with high systematic error (i.e. low Dice index for the pCT). The spatial distribution of improvements was heterogeneous, where Figure 1b shows the greatest average reduction in systematic errors for the cranial- anterior part of the rectum. Dosimetric evaluation found significant improvements in the DVH-range of 52.5 to 65 Gy, seen by the enlarged region of Figure 2, but not in EUD or D5% as compared to the pCT.

Conclusion Shrinkage estimation for the rectum shape was feasible, and resulted in higher similarities to the patient mean rectum. The method has potential to increase the accuracy in RT of deformable organs with applications in toxicity modelling and plan optimisation.

Symposium: Automatic contouring: Metrics and clinical evaluation approaches

SP-0365 Overview of AI-based methods for automatic contouring of OARs and tumours C. Brouwer 1 1 University of Groningen, University Medical Center Groningen, Department of Radiation Oncology, Groningen, The Netherlands Abstract Text Deep learning for automated segmentation (automatic contouring) in radiotherapy is or will soon be the new standard for segmentation of organs at risk (OARs) and target volumes in radiotherapy images. This talk will provide an overview of presented methods focusing on its application: the nominal workflow, the adaptive workflow and retrospective studies. In addition, the application in ESTRO member clinics will be presented based on a 2020 survey.

SP-0366 Contour similarity metrics and clinical usability M. Gooding 1 1 Mirada Medical Ltd, Science and Medical Technology, Oxford, United Kingdom