ESTRO 2024 - Abstract Book

S3144

Physics - Autosegmentation

ESTRO 2024

We demonstrate that the AI segmented structures show favorable metrics, are subject to major corrections in only 3 out of 110 structures, and exhibit a strong correlation between mean dose to AI segmented and manually delineated structures. These results show that AI can be a safe and time-efficient tool for the segmentation of pelvic bone substructures. Our results indicate that dose distributions will be unaffected even if the AI-segmented structures are not corrected. We are currently implementing the algorithm in clinical practice.

Keywords: Anal Cancer, Pelvic bone structures, AI

2967

Digital Poster

A combined automatic MR OAR contouring and synthetic CT solution for prostate MR-only radiotherapy

Jonathan J Wyatt 1,2 , Sandeep Kaushik 3,4 , Cristina Cozzini 3 , Bernadett Kolozsvári 5 , Borbála Deák-Karancsi 5 , Rachel A Pearson 2,1 , Steven Petit 6 , Marta Capala 6 , Juan A Hernandez-Tamames 7 , Katalin Hideghéty 8 , Ross J Maxwell 1 , László Ruskó 5 , Florian Wiesinger 3 , Hazel M McCallum 1,2 1 Newcastle University, Translational and Clinical Research Institute, Newcastle, United Kingdom. 2 Newcastle upon Tyne Hospitals NHS Foundation Trust, Northern Centre for Cancer Care, Newcastle, United Kingdom. 3 GE Healthcare, Research, Munich, Germany. 4 University of Zurich, Department of Quantitative Biomedicine, Zurich, Switzerland. 5 GE Healthcare, Research, Budapest, Hungary. 6 Erasmus MC Cancer Institute, Department of Radiotherapy, Rotterdam, Netherlands. 7 Erasmus MC, Department of Radiology and Nuclear Medicine, Rotterdam, Netherlands. 8 University of Szeged, Department of Oncotherapy, Szeged, Hungary MR is preferred for contouring in prostate radiotherapy due to its superb soft-tissue contrast(1). MR cannot be directly used for radiotherapy planning which has led to the development of synthetic CT (sCT) solutions to facilitate MR-only radiotherapy(2). This improves treatment efficiency by omitting a planning CT appointment, as well as improving treatment accuracy by removing the MR-CT registration uncertainty(3). In addition, the improved MR image quality has motivated MR-based Deep Learning automatic contouring algorithms, which could further improve the consistency and efficiency of contouring. A combined automatic MR OAR contouring and sCT solution for the pelvis has been developed, with each component separately validated. Clinicians rated 81% of automatic MR OAR contours as clinically acceptable(4) and mean PTV dose differences of the sCT to CT were ≤ 0.5%(5). The aim of this study was to evaluate the combined automatic MR OAR contouring and sCT workflow against manual MR contours on a MR-CT workflow for prostate MR-only radiotherapy. Purpose/Objective:

Material/Methods:

10 prostate radiotherapy patients received planning MR and CT scans in the radiotherapy planning position. Rapid (1 minute) zero echo time and T2-weighted turbo spin echo sequences were acquired. sCTs were generated from the zero echo time sequence by a Deep Learning 2D convolutional neural network(5,6). A set of 2D and 3D convolutional neural networks were used to produce automatic contours for the bladder, bowel bag, femoral heads, rectum, penile