ESTRO 2024 - Abstract Book

S4090

Physics - Inter-fraction motion management and offline adaptive radiotherapy

ESTRO 2024

F0 models were comparable to BM, however, PS noBM

F0 exhibited a wider

The median DSC and HDs values for PS noBM range of values and more outliers. PS noBM

F0-4 outperformed BMs in all metrics. Figure 2 shows the outcomes of the

clinical evaluation.

PS BM F0 predictions with an average grade of 1.02 could be used right away for treatment adaptation most of the time, while BM delineations with an average grade of 1.54 required more corrections. PS noBM F0 contours received an average grade of 2.36 meaning that they frequently required minor or major corrections. Among the organs studied, the aorta, bowel, duodenum, and spinal canal benefited the most from PS training, whereas improvements in the kidneys, liver, and stomach were moderate.

Conclusion:

In this study, we demonstrated the advantages of personalized segmentation models in fractionated MRgRT of the abdomen region compared to conventional BMs. Particularly, PS models generated through fine-tuning the BM with individual patient data performed the best, while training from scratch showed worse performance than the BMs. Training with additional fractions further improved the models, however this approach is the most challenging to implement clinically. The physician assessment showed that fine-tuning BMs with only the planning MRI generates delineations that, in most cases, can be used directly for plan adaptation, and only few require major corrections.

Keywords: MRgRT, personalized models, auto-segmentation

References:

[1] S. Fransson, D. Tilly, and R. Strand. Patient specific deep learning based segmentation for magneticresonance guided prostate radiotherapy. Physics and Imaging in Radiation Oncology, 23:38–42,2022. [2] M. Kawula, I. Hadi, L. Nierer, M. Vagni, D. Cusumano, L. Boldrini, L. Placidi, S. Corradini,C. Belka, G. Landry, et al. Patient-specific transfer learning for auto-segmentation in adaptive0.35 T MRgRT of prostate cancer: a bi-centric evaluation. Medical Physics, 50(3):1573–1585,2023. [3] Z. Li, W. Zhang, B. Li, J. Zhu, Y. Peng, C. Li, J. Zhu, Q. Zhou, and Y. Yin. Patient-specific dailyupdated deep learning auto-segmentation for MRI-guided adaptive radiotherapy. Radiotherapy andOncology, 177:222–230, 2022.2

Acknowledgments: Wilhelm-Sander-Stiftung (2019.162.2)