ESTRO 2021 Abstract Book

S1418

ESTRO 2021

Conclusion The suggested model ensemble shows excellent segmentations of multiple OAR structures in the pelvis region, and achieves improved results even for highly complex structures like bowel, small and large bowel (figure 1) . Exploring ways to improve the segmentations of small bowel, large bowel, and rectum seems to be the natural next step. At the current state, our proposed model might be used to accelerate the manual segmentation process or, for some OAR, fully automate the delineation task.

PO-1692 Principal curve-based and AI-based method for ultrasound prostate delineation T. Peng 1 , C. Tang 2 , J. Cai 1 1 The Hong Kong Polytechnic University, Health Technology and Informatics, Hong Kong, Hong Kong (SAR) China; 2 Taizhou People’s Hospital, Department of Medical Imaging, Taizhou, China Purpose or Objective Delineating Organs-At-Risk (OAR) is essential for radiotherapy planning and quantitative imaging workflows. However, manual contouring is highly laborious and time-consuming, and prone to variability and poor reproducibility. To address these issues we design an accurate ultrasound prostate delineation method combining an improved greedy closed principal curve method with an improved differential evolution-based machine learning method. Materials and Methods A total of 50 brachytherapy patients were used, which was randomly divided into three groups, 30 for training,