S2231
Physics - Intra-fraction motion management and real-time adaptive radiotherapy
ESTRO 2026
Zhang DM, et al. Nature Communications. 2021;12(1).[3] Mayinger M, et al. Radiotherapy and Oncology. 2020;152:203-207. [4] O Akdag, et al. Physics in Medicine and Biology. 2022;67(6):065003- 065003 [5] Paul, Dong B, et al. Medical Physics. 2020;47(12):6440-6449.[6] Snoek L, et al. Scientific Data. 2021;8:85.[7] Blöcker T, et al. Physics in Medicine and Biology. 2024;70(1) Keywords: MRgRT, STAR, VT Digital Poster Highlight 2497 Accelerating care and enhancing precision: A same-day, MRI-only simulation workflow for SRS of brain metastases in preparation of trial NCT07132190 Marvin Kinz 1,2 , Daniel C. Miller 1 , Thomas Ciavattone 1 , Evangelia Kaza 1 , Scott Friesen 1 , Cassandra L. Bullens 1 , Maria Czerminska 1 , Jürgen Hesser 3,4 , Atchar Sudhyadhom 1 , Christian V. Guthier 5 , Jeremy Bredfeldt 1 , Ayal A. Aizer 1 , Kamal Singhrao 1 1 Department of Radiation Oncology, Brigham and Women's Hospital, Dana-Farber Cancer Institute, Harvard Medical School, Boston, USA. 2 Department of Physics and Astronomy, Interdisciplinary Center for Scientific Computing (IWR), Heidelberg University, Heidelberg, Germany. 3 Department of Physics and Astronomy, Interdisciplinary Center for Scientific Computing (IWR), Central Institute for Computer Engineering (ZITI), CZS Heidelberg Initiative for Model- Based AI, Heidelberg University, Heidelberg, Germany. 4 Mannheim Institute for Intelligent Systems in Medicine (MIISM), Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany. 5 Department of Radiation Oncology, Dana-Farber Cancer Institute, Harvard Medical School, Boston, USA Purpose/Objective: The efficacy of stereotactic radiosurgery (SRS) for brain metastases can be undermined by tumor growth and soft tissue displacement, which can occur between simulation imaging and treatment [1]. Additionally, larger brain metastasis volumes are associated with decreased overall survival and significantly increased risk of grade 3, 4, or 5 neurotoxicity [2]. We developed a same-day MRI-only simulation workflow for Linac- based SRS to directly address these challenges. Our goal is to accelerate treatment initiation, reduce delay between planning and delivery, eliminate multimodality registration errors [3], and enhance therapeutic precision, ultimately enabling smaller treatment margins and improving the standard of care. Material/Methods: We established an integrated workflow where patients undergo a single, high-resolution MRI simulation in the
Additionally, SAM2 segmentations of the blood pool were compared to manual segmentations using Dice similarity coefficient (DSC), 50th and 95th percentile Hausdorff (HD50, HD95), and Euclidean center distances (ECD). Pairwise inter-observer variability (IOV) was estimated from segmentations of 4 observers. Results: SAM2 achieved segmentation accuracy (median ± interquartile range (IQR) DSC: 0.93 ± 0.03, HD50: 2.4 ± 0.9 mm, HD95: 4.8 ± 1.7 mm, ECD: 1.8 ± 0.6 mm), comparable to the mean pairwise IOV (DSC: 0.91 ± 0.05, HD50: 2.5 ± 1.1 mm, HD95: 5.3 ± 1.8 mm, ECD: 1.9 ± 0.9 mm).
The online LSTM achieved the lowest RMSE across all testing cohorts, outperforming the offline LSTM (p = 0.011), linear regression (p < 0.001), and no prediction (p < 0.001) for 360 ms (Table 2). LSTM inference and retraining times averaged 0.7 ms and 34 ms, respectively. Conclusion: In-silico LSTM and SAM2 results in healthy volunteers and non-STAR patients indicate the potential for reliable localization and motion prediction in MRIgRT STAR applications. References: [1] Marcin Miszczyk, et al. Heart Rhythm. 2024.[2]
Made with FlippingBook - Share PDF online