S2272
Physics - Intra-fraction motion management and real-time adaptive radiotherapy
ESTRO 2026
Digital Poster 4616
Initial Clinical Experience of Comprehensive Motion Management for Magnetic Resonance- Guided Stereotactic Body Radiotherapy in Pancreatic Cancer Amy Tien Yee Chang 1 , Bin Yang 2 , Chen-Yu Huang 2 , Cheuk Yan Ng 2 , Ka Ki Lau 2 , Jing Yuan 3 , Oi Lei Wong 3 , Kin Yin Cheung 2 , Siu Ki Yu 2 , Stephen Chun Key Law 1 , Darren Ming Chun Poon 1 1 Comprehensive Oncology Centre, Hong Kong Sanatorium & Hospital, Hong Kong, China. 2 Medical Physics Department, Hong Kong Sanatorium & Hospital, Hong Kong, China. 3 Research Department, Hong Kong Sanatorium & Hospital, Hong Kong, China Purpose/Objective: Magnetic resonance-guided stereotactic body radiotherapy (MRgSBRT) using a 1.5 T MR-Linac provides superior soft-tissue visualization and adaptive treatment capability. However, pancreatic tumors remain among the most challenging SBRT targets due to complex respiratory-induced motion and intrafractional organ drift. Comprehensive Motion Management (CMM), integrating automatic gating and real-time tumor tracking, offers a novel strategy to address these motion uncertainties. This study reports the initial clinical experience of MRgSBRT for pancreatic cancer using CMM with true tumor tracking. Material/Methods: Ten patients with locally advanced or metastatic pancreatic cancer were treated with MRgSBRT using CMM. Abdominal compression was applied in eight patients, with three patients receiving multiple lesion treatment. PTV margins of 3–5mm were applied to CTV, with envelope structure margins of 3–5mm for gating motion ranges. Prescription doses to PTV ranged from 33 to 40Gy in five fractions, with GTV boost doses of 40–50Gy. Volumetric Overlapping Criterion (VOICE) setting was initially set to 100%. Baseline shift tolerance was set at 5mm (3mm in two patients). Fifty-three treatment sessions were analyzed for duty cycle, beam-on time, motion amplitude, and baseline shift corrections. Results: Median beam-on time was 30 minutes (range: 15– 69min), with median duty cycle of 56% (19.3–85.3%). Duty cycles <50% and <30% occurred in 34% and 6% of sessions (18/53 and 3/53), respectively. Baseline shift correction was required in 34% of sessions (18/53), with ≥ 3 corrections in only 4% (2/53). Mean final baseline shift displacements were 0.10mm (left-right), 0.51mm (superior-inferior), and − 1.50mm (anterior- posterior), with multidirectional shifts in 30% of sessions. VOICE settings were adjusted in 13% of sessions: 19% (10/53) set to 95%, 11% (6/54) adjusted
Conclusion: DL-based auto-segmentation models showed promising performance for real-time liver localization. TransMorph trained on synthetic (DRR) data only, generalized effectively when validated on real X-rays. Evaluation on additional real X-rays will follow shortly. References: [1] T. Akinci D’Antonoli et al., ‘TotalSegmentator MRI: Robust Sequence-independent Segmentation of Multiple Anatomic Structures in MRI’, Radiology, vol. 314, no. 2, p. e241613, 2025, doi: 10.1148/radiol.241613.[2] N. Ravi et al., ‘SAM 2: Segment Anything in Images and Videos’, p. F. Isensee et al., ‘nnU-Net Revisited: A Call for Rigorous Validation in 3D Medical Image Segmentation’, p. arXiv:2404.09556, 2024, doi: 10.48550/arXiv.2404.09556.[4] E. arXiv:2408.00714, 2024, doi: 10.48550/arXiv.2408.00714.[3] Lombardo et al., ‘Patient-Specific Deep Learning Tracking Framework for Real-Time 2D Target Localization in Magnetic Resonance Imaging-Guided Radiation Therapy’, Int J Radiat Oncol Biol Phys, 2024, doi: 10.1016/j.ijrobp.2024.10.021. Keywords: X-ray, markerless, tracking
Made with FlippingBook - Share PDF online