ESTRO 2021 Abstract Book

S772

ESTRO 2021

Conclusion This is the first systematic comparison of an auto-contouring algorithm’s performance on a standardized OAR dataset in RT. Our results need to be confirmed on a larger patient cohort, but indicate that the MVision algorithm did very well since the size of identified differences generally was small both with respect to clinical and to reference volumes. However, to increase the use of AI-algorithms for clinical studies and research, retraining them on standardized reference datasets has the potential to improve performance further. PD-0930 Comparison of automated segmentation techniques for magnetic resonance images of the prostate M. Pepa 1 , J.L. Isaksson 1 , M. Zaffaroni 1 , P.E. Summers 2 , G. Marvaso 1,3 , G. Lo Presti 4 , S. Raimondi 4 , S. Gandini 4 , S. Volpe 1,3 , D.P. Rojas 1 , D. Zerini 1 , Z. Haron 5 , P. Pricolo 6 , S. Alessi 6 , F.A. Mistretta 7 , S. Luzzago 7 , F. Cattani 8 , O. De Cobelli 7,3 , E. Cassano 9 , M. Cremonesi 10 , M. Bellomi 6,3 , R. Orecchia 11 , G. Petralia 6,3 , B.A. Jereczek-Fossa 1,3 1 IEO European Institute of Oncology IRCCS, Division of Radiation Oncology, Milan, Italy; 2 IEO European Institute of Oncology IRCCS, Division of Radiology, MIlan, Italy; 3 University of Milan, Department of Oncology and Hemato-Oncology, Milan, Italy; 4 IEO European Institute of Oncology IRCCS, Molecular and Pharmaco- Epidemiology Unit, Department of Experimental Oncology, Milan, Italy; 5 National Cancer Institute, Radiology Department, Putrajaya, Malaysia; 6 IEO European Institute of Oncology IRCCS, Division of Radiology, Milan, Italy; 7 IEO European Institute of Oncology IRCCS, Division of Urology, Milan, Italy; 8 IEO European Institute of Oncology IRCCS, Medical Physics Unit, Milan, Italy; 9 IEO European Institute of Oncology IRCCS, Division of Breast Radiology, Milan, Italy; 10 IEO European Institute of Oncology IRCCS, Radiation Research Unit, Milan, Italy; 11 IEO European Institute of Oncology IRCCS, Scientific Direction, Milan, Italy Purpose or Objective The contouring of regions of interest (ROIs) is a crucial step in the radiomic workflow and is often both time- consuming and prone to intra- and inter-observer variability. The present investigation aims at comparing different strategies for automatically segmenting the prostate in T2-weighted magnetic resonance images The study involved a set of 100 patients with diagnosis of prostate adenocarcinoma who had undergone pre- surgical multi-parametric MRI and prostatectomy in our Institution between 2014 and 2018. Axial T2-weighted MR images of the pelvis were exported from the institutional imaging archive and the prostate was manually contoured by a junior radiologist with less than 5-year experience (ZH) and subsequently checked by three expert radiologists (PP, SA, GP) to ensure their correctness and define a ground truth. The prostate was then automatically contoured using six different methods: (1) a commercial package (SyngoVia, Siemens Healthcare, Erlangen Germany), (2) an ad hoc multi atlas-based algorithm (Raystation 9B, RaySearch Laboratories, Stockholm Sweden) and (3-6) four U-net based deep learning (DL) networks. The DL networks (U- net, Transfer Learning (TL), Generative Adversarial Network (GAN), and EfficientDet3D (ED3D)) differed in (MRIs) of the male pelvis. Materials and Methods