ESTRO 2025 - Abstract Book

S4381

RTT - Treatment planning, OAR and target definitions

ESTRO 2025

Conclusion: For organs, where the tissue density differs significantly from their surroundings (air-soft tissue, bone-soft tissue) the contours can be delineated with good agreement and high quality. For small-volume organs, relatively large deviations were found, therefore improvement in contouring software is recommended to reach better consistency.

Keywords: artificial intelligence, organ at risk, contouring

3473

Digital Poster Initial evaluation of AI-contoured OARs in CBCT-guided online adaptive radiotherapy for focal bladder and cervical cancer Karin Goudschaal 1,2 , Sana Azzarouali 3,2 , Jorrit Visser 1,2 , Maryam Afifah 3,2 , Anouk Maasland 3,2 , Laurien Daniels 1,2 , Arjan Bel 1,2 , Duncan den Boer 3,2 1 Radiation Oncology, Amsterdam UMC location University of Amsterdam, Amsterdam, Netherlands. 2 Cancer Center Amsterdam, Cancer Treatment and Quality of Life, Amsterdam, Netherlands. 3 Radiation Oncology, Amsterdam UMC location Vrije Universiteit Amsterdam, Amsterdam, Netherlands Purpose/Objective: In an online adaptive radiotherapy (oART) workflow, short treatment times and high accuracy of delineated contours are crucial for treatment quality and broader implementation. Artificial intelligence (AI)-driven automatic contouring has the potential to enhance precision and efficiency, reducing intrafraction motion [1-5]. The aim of this initial study is to evaluate the quality of AI-contoured organs-at-risk (OARs) in CBCT-guided oART for focal bladder and cervical cancer. This study validates AI-contoured OARs against our clinical delineation practices, as this is essential for embedding this tool into our online adaptive Radiation Therapist (RTT)-only workflow. Material/Methods: We used the online Ethos 2.0 test environment (Emulator, version 2.0, Varian a Siemens Healthineers Company, USA) to retrospectively simulate target deformation and assess the accuracy of AI-contoured OARs, focusing on influencer structures (i.e. contours that play a significant role in target propagation; in this case, the bladder and rectum). The bowel bag was included as an important OAR to limit toxicity, and the femoral heads were used as benchmarks. Five planning CTs (pCTs) from three bladder cancer patients (two with full bladder pCTs and one with both empty and full bladder pCTs) and one cervical cancer patient, treated with oART at our institute between November 2021 and November 2024, were analyzed. AI-contoured OARs were compared with the manual delineations used in our pre treatment workflow, employing the DICE similarity coefficient (DSC) and Hausdorff distance (HD) [6,7]. Results: AI contouring achieved a median DSC of 0.92–1.00 for the different OARs, indicating strong similarity with manual delineations (Figure 1A), with the lowest being the rectum. The median HD for the different OARs ranged from 1.8 to 6.0 mm, showing large boundary distances between AI-generated and manual contours (Figure 1B), except for the bowel bag (1.0 mm).