ESTRO 2025 - Abstract Book

S2512

Physics - Autosegmentation

ESTRO 2025

Conclusion: This study demonstrated that positional guide played key a role in DLAS for VS, enhancing segmentation robustness and efficacy, making it clinically viable for GK-SRS in a CPU environment. PG can be integrated into various deep learning networks through channel-wise concatenation.

Keywords: Vestibular schwannoma, auto-segmentation, postion

References: 1. Carlson, M. L. & Link, M. J. Vestibular Schwannomas. New England Journal of Medicine 384 , 1335–1348 (2021). 2. Kujawa, A. et al. Deep learning for automatic segmentation of vestibular schwannoma: a retrospective study from multi-center routine MRI. Front Comput Neurosci 18 , (2024). 3. Shapey, J. et al. Segmentation of vestibular schwannoma from MRI, an open annotated dataset and baseline algorithm. Sci Data 8 , 8–13 (2021). 4. Cardoso, M. J. et al. MONAI: An open-source framework for deep learning in healthcare. (2022) doi:https://doi.org/10.48550/arXiv.2211.02701. 5. Hatamizadeh, A. et al. Swin UNETR: Swin Transformers for Semantic Segmentation of Brain Tumors in MRI Images. (2022).

3205

Digital Poster Vertebrae Growth Tracking in Paediatric CSI Patients: Correlating CT and MRI measurements for long term late effects assessment Kartik Kumar 1,2 , Adam Yeo 1,2,3 , Tomas Kron 1,2,3 , Rick Franich 1,2 1 School of Science, RMIT University, Melbourne, Australia. 2 Physical Sciences Department, Peter MacCallum Cancer Centre, Melbourne, Australia. 3 Sir Peter MacCallum Department of Oncology, University of Melbourne, Melbourne, Australia Purpose/Objective: Cranio-spinal irradiation (CSI) paediatric patients undergo MRI follow-up to monitor treatment-related late effects, such as growth retardation and scoliosis. AI advancements offer potential to automate vertebral growth measurement using longitudinal MRI data. A pair of CT and MR images is acquired for CSI, followed by MR imaging for post-RT follow-up. However, understanding the correlation between vertebral height measurements on CT and MR is essential for accurately tracking growth post-radiotherapy. This study aims to develop an automated framework to measure and track the growth of each vertebra on T2-weighted MR images and compare to those obtained from CT images. Material/Methods: An automated framework was developed to extract anterior and posterior vertebral heights using two U-Net-based segmentation networks for CT and T2w-MR. The networks were trained on 250 CT (170 Adult, 70 Paediatric) and 240 MR (210 Adult, 30 Paediatric) training pairs to produce binary vertebrae masks. Post-processing classified individual vertebrae and extracted landmarks for height measurements (Figure 1). The study analysed 21 paired CT and T2 weighted MR scans from the same paediatric patients (aged 5–12 years), with each pair acquired <30 days apart. CT images had a resolution of 1.17×1.17 mm, slice thickness of 2–3 mm, and MR had a Spin-Echo (SE) pulse sequence with 0.45×0.45 mm resolution and 3.3 mm slice thickness. A total of 540 vertebral bodies (C4–L5) were measured.