S2199
Physics - Intra-fraction motion management and real-time adaptive radiotherapy
ESTRO 2026
Riboldi 7 , Denis Dudá š 1 , Hilary Byrne 8,9 , Lorenzo Placidi 10 , Marco Fusella 11 , Michael Jameson 8 , Miguel Palacios 12 , Paul Keall 13 , Matteo Maspero 6,14 , Christopher Kurz 1 , Guillaume Landry 1,15 1 Department of Radiation Oncology, LMU University Hospital, Munich, Germany. 2 Department of Radiation Oncology, Catharina Hospital, Eindhoven, Netherlands. 3 Department of Radiation Oncology, Radiation Oncology Key Laboratory of Sichuan Province, Sichuan Cancer Hospital & Institute, Chengdu, China. 4 Department, LMU University Hospital, Munich, Germany. 5 Medical Physics Unit, Mater Olbia Hospital, Olbia, Italy. 6 Department of Radiotherapy, University Medical Center Utrecht, Utrecht, Netherlands. 7 Department of Medical Physics, Ludwig-Maximilians- Universität München, Garching, Germany. 8 GenesisCare, St Vincent’s Hospital, Sydney, Australia. 9 Sydney School of Health Sciences, Faculty of Medicine and Health, The University of Sydney, The University of Sydney, Australia. 10 Department of Diagnostic Imaging, Oncological Radiotherapy and Hematology, Fondazione Policlinico Universitario Agostino Gemelli, IRCCS, Rome, Italy. 11 Department of Radiation Oncology, Abano Terme Hospital, Abano Terme Veneto, Italy. 12 Department of Radiation Oncology, Amsterdam UMC location Vrije Universiteit Amsterdam, Amsterdam, Netherlands. 13 Sydney School of Health Sciences, Faculty of Medicine and Health, The University of Sydney, Sydney, Australia. 14 Computational Imaging Group for MR Diagnostics & Therapy, University Medical Center Utrecht, Utrecht, Netherlands. 15 Bavarian Cancer Research Center (BZKF), LMU University Hospital, Munich, Germany
and 50 labeled cases were made public. The hidden test set included 8 labeled cases for the public preliminary evaluation phase and 50 for the final evaluation.Submissions were executed on provided hardware and evaluated using six metrics. Dice similarity coefficient (DSC), Hausdorff-distance 95 (HD95), Euclidean center distance (ECD), mean average surface distance (MASD), a metric estimating the relative reduction of D98 target dose due to errors in MLC-tracking (RD98), and runtime per frame (TPF). Submissions were ranked per metric, with final leaderboard positions determined via Rank-Then- Mean. The Top-5 submission reports were peer- reviewed and algorithms were made open source.
Purpose/Objective: Magnetic resonance imaging (MRI)-guided
radiotherapy (MRIgRT) combines MRI with radiation therapy linear accelerators (MRI-linacs), allowing motion monitoring with CineMRI. Current systems typically rely on deformable image registration or template matching to track tumors for gated beam delivery, but there is growing interest in AI-driven approaches. These could facilitate multi-leaf collimator (MLC) tracking-based delivery, potentially enhancing duty cycles and throughput. (1,2) Material/Methods: TrackRAD2025 (3), introduced at ESTRO 2025 and concluded at MICCAI 2025, was organised to foster and evaluate solutions for target tracking. Challenge participants were tasked with developing algorithms to propagate segmentation masks from initialisation frames across subsequent frames.The dataset (4) contained 2D sagittal CineMRIs acquired at 0.35T (MRIdian, ViewRay) and 1.5T (Unity, Elekta) MRI-linacs from six institutions across three continents. 585 sequences (>2 million frames) were collected. Experts annotated targets in 108 sequences. 477 unlabeled
Results: TrackRAD2025 achieved substantial participation, with 166 registrations and 100 submissions to the preliminary testing phase. The final testing phase received 24 submissions from 15 teams, of which 14 qualified for the final leaderboard.
Made with FlippingBook - Share PDF online