S1591
Physics - Autosegmentation
ESTRO 2026
segmentation tools for BM in SRS remains insufficiently validated, particularly regarding detection accuracy and dosimetric relevance. Material/Methods: All patients treated with robotic SRS for BM in 2024 were retrospectively identified, and their contrast- enhanced T1-weighted MRIs were subsequently analyzed using the AI-based Cranial Tumor Segmentation algorithm (RT Elements 4.5, Brainlab AG, Munich, Germany, v4.5.0.273). Clinically, the gross tumor volume (GTV) was delineated by an experienced radiosurgeon, and the planning target volume (PTV) was defined by isotropically expanding the GTV by 1 mm, serving as the reference PTV. AI-generated GTVs were compared to the reference PTVs using the Dice Similarity Coefficient (DSC) after isotropic expansion by 0.5, 1.0, 1.5, and 2.0 mm. To assess the clinical relevance of volumetric deviations, dose coverage of the AI-generated structures (with corresponding margins) and lesion detection rate were evaluated based on the delivered treatment plans. Results: Sixty patients with 180 BM were analyzed. The AI algorithm detected 152 of 180 lesions (sensitivity = 0.84), with smaller metastases more frequently missed (median: 0.08 cm³; range: 0.02-0.62 cm³). Median AI- segmented volume (0.12 cm³; range 0.01–25.65 cm³) was smaller than the manual GTV (0.43 cm³; range 0.02–42.53 cm³). The highest median DSC was observed with 1.5 mm (0.81) and 2.0 mm (0.80) margins. A 1.5 mm margin achieved a median dose coverage of 97.37 %, comparable to the reference PTV coverage (99.26 %). Overall, AI-generated contours demonstrated high concordance with manual delineations, with a 0.5 mm margin to the GTV or a 1.5 mm margin to approximate the PTV providing the best balance between geometric and dosimetric agreement.
identify lesions not detected by neuroradiologists was not evaluated in this study. The generally smaller AI contours compared to manual delineations could be compensated by applying suitable margins to ensure adequate target coverage. While the algorithm may enhance contouring efficiency and standardization, manual verification remains essential to ensure clinical accuracy. References: [1] Aizer, A. et al.;Stereotactic radiation versus hippocampal avoidance whole brain radiation in patients with 5-20 brain metastases: A multicenter, phase 3 randomized trial;Journal of Clinical Oncology; June 2025; DOI: 10.1200/JCO.2025.43.16_suppl.2011[2] Madhugiri, V.S. et al.;Early experience with an artificial intelligence-based module for brain metastasis detection and segmentation; Journal of Neuro- Oncology; October 2024; DOI: 10.1007/s11060-024- 04851-8 Keywords: Brain Metastases, stereotactic radiosurgery Clinical and automated heart segmentation and dose-volume metric variation for lung and breast radiotherapy across the Australian Cancer Data Network Lois C Holloway 1,2 , Fahim Alam 2,1 , Phillip Chlap 2,1 , Robert Finnegan 3 , Vicky Chin 1,2 , Sasha Barisic 1,2 , Amir Anees 2,1 , Matthew Field 2 , Michael Bailey 4 , Jonathan Sykes 5,3 , Simon Ashworth 5 , Stuart Greenham 6 , Nicholas Hardcastle 7 , Catherine Lawford 7 , Prabhakar Ramachandran 8 , Elizabeth Claridge 9,3 , Anthony Espinoza 1,2 , Senthilkumar Gandhidasan 4 , Kirsty Stuart 5,10 , Verity Ahern 5,10 , Joerg Lehmann 11,3 , Jane Ludbrook 12 , David Thwaites 3 , Shalini Vinod 1,2 , Geoff Delaney 1,2 1 Liverpool and Macarthur Cancer Therapy Centre and Ingham Institute, South Western Sydney Local Health Digital Poster Highlight 4078 District, Sydney, Australia. 2 Faculty of Medicine, University of New South Wales, Sydney, Australia. 3 Institute of Medical Physics, University of Sydney, Sydney, Australia. 4 Illawarra Cancer Care Centre, Illawarra Shoalhaven Local Health District, Wollongong, Australia. 5 Crown Princess Mary Cancer Centre and Blacktown Cancer and Haematology Centre, Western Sydney Local Health District, Sydney, Australia. 6 Mid North Coast Cancer Institute, Mid North Coast Local Health District, Coffs Harbour, Australia. 7 Department of Physical Sciences, Peter MacCallum Cancer Centre, Melbourne, Australia. 8 Department of Radiation Oncology, Princess Alexandra Hospital, Brisbane, Australia. 9 Radiation Oncology, Chris O'Brien Lifehouse, Sydney, Australia. 10 Faculty of Medicine, University of Sydney, Sydney, Australia. 11 Radiation
Conclusion: The evaluated AI-based segmentation tool
demonstrated good agreement with manual contours, particularly for larger BM. Although smaller lesions were occasionally missed, the potential of AI to
Made with FlippingBook - Share PDF online