S2230
Physics - Intra-fraction motion management and real-time adaptive radiotherapy
ESTRO 2026
Czech, (PTC), Prague, Czech Republic. 10 Department of Medical Physics, Faculty of Physics, Munich, Germany. 11 Department of Radiation Oncology, Universitätsklinikum Erlangen, Friedrich-Alexander- Universität Erlangen-Nürnberg (FAU), Erlangen, Germany Purpose/Objective: Stereotactic arrhythmia radioablation (STAR) aims to eliminate refractory reentrant ventricular tachycardia by homogenizing electrophysiological substrate using single-fraction irradiation [1,2]. Magnetic resonance imaging (MRI)-guided radiotherapy (MRIgRT) may offer superior precision for treatments, but due to system latencies exceeding 300 ms [3], motion prediction is required [4]. This study investigates artificial intelligence-based localization and motion prediction methods for MRI-linac derived cardiorespiratory motion traces from cine-MRI. Material/Methods: A long short-term memory (LSTM) network was initially trained on synthetic motion trajectories generated from surrogate data of the Amsterdam Open MRI collection [6]. Subsequently, the network was further trained on 2D cine-MRI-derived trajectories from 11 healthy volunteers, acquired at 11 Hz on a 0.35T MRI- linac in sagittal and coronal orientations. Time- resolved segmentations of the blood pool in the left ventricle were retrospectively generated using SAM2 [7] and split into cardiac quadrants to extract 2D trajectories (Figure 1). The LSTM was configured to predict the next 360 ms, based on the previous 9 s. Between predictions, the LSTM was fine-tuned in real- time (online LSTM) and compared to its static counterpart (offline LSTM). Real-time linear regression and a baseline without prediction (last known position) were implemented for comparison. Prediction accuracy was evaluated with root mean square error (RMSE) for 2 cohorts: internal lung cancer patients with additional cardiac scans at the 0.35T MRI-Linac (cohort 1), and external healthy volunteers scanned at a 1.5T MRI-Linac (cohort 2). A post-hoc nemenyi-friedman test assessed statistical significance.
of 30 patients treated for spinal or sacral lesions. For each patient, all triggered kV frames acquired during treatment were analyzed offline. Results: Phantom experiments demonstrated sub-millimetric repeatability, with mean displacement of -0.2 mm ± 0.5 mm (horizontal) and -0.0 mm ± 0.3 mm (vertical) across repeated acquisitions.In the clinical cohort, a total of 150 fractions and 5400triggered images were analyzed. The mean absolute intrafraction displacement was -0.1 mm ± 1.1 mm horizontally and - 0.1 mm ± 1.0 mm vertically, with a mean 2D displacement magnitude of 1.2 mm ± 0.9 mm. Maximum observed displacement reached 6.9 mm. In several cases, transient shifts > 2 mm in at least 3 consecutive images occurred despite image-guided setup and immobilization. Conclusion: This offline workflow enables patient-specific, angle- resolved intrafraction motion verification using routinely acquired per-fraction triggered kV images, without modifying clinical workflow or acquisition protocols. The approach provides actionable quantitative data that can support margin evaluation, immobilization QA, and future adaptive strategies. Integration into routine quality assurance may help reinforce geometric precision, particularly in high- dose-gradient treatments such as spine SBRT. Keywords: Intrafraction motion; IGRT; Quality assurance Real-time localization and motion prediction for MRI-guided stereotactic arrhythmia radioablation Nicolas Mühlschlegel 1,2 , Axel Hengesse 1 , Elia Lombardo 1 , Tom Blöcker 1 , Rabea Klaar 3,4 , Manon Aubert 5 , Christianna I Papadopoulou 1 , Alexandre Pfeffer 6 , Guillaume Poirot 7 , Martin Domansky 8,9 , Marco Riboldi 10 , Claus Belka 1,2 , Stefanie Corradini 1,11 , Christopher Kurz 1 , Martin Fast 5 , Guillaume Landry 1,2 1 Department of Radiation Oncology, LMU University Hospital, Munich, Germany. 2 Bavarian Cancer Research Center, (BZKF), Munich, Germany. 3 Department of Radiology, LMU University Hospital, Munich, Germany. 4 Comprehensive Pneumology Center (CPC-M), Member of the German Center for Lung Research (DZL), Munich, Germany. 5 Department of Radiotherapy, University Medical Center Utrecht, Utrecht, Netherlands. 6 Greater Paris University Hospitals - APHP, INSERM UMR-S 942, Paris, France. 7 Department of Cardiothoracic Imaging, Hôpital Cardiologique du Haut-Lévêque, CHU de Bordeaux, Pessac, France. 8 Department of Oncology, 2nd Faculty of Medicine, Charles University Prague and Motol, Prague, Czech Republic. 9 Proton Therapy Center Proffered Paper 2246
Made with FlippingBook - Share PDF online