[Poster] Deep Learning-based dose distribution correction of Electron Return Effect in MR-guided radiotherapy

Alexandre Hakimi^a, Eric Fadel^a, François Smekens^a, Clément Chevillard^b, François Husson^a

Introduction: MR-guided radiotherapy systems (e.g. Elekta Unity) enable online adaptive workflows, requiring fast and reliable secondary dose calculation (SDC). Monte Carlo (MC) methods accurately model the Electron Return Effect1 (ERE), but their statistical uncertainty limits their use for SDC. In contrast, Collapsed Cone Convolution (CCC) is fast,only accounts for the effects of magnetic field in homogeneous media using a specific warped kernels, leading to false positives when compared to MC-based TPS doses. In this work, we propose a deep learning–based correction of CCC dose distributions, aiming to combine the speed of CCC with the accuracy of MC in the presence of a magnetic field.

Résultats: The performance of the model was evaluated using 3D gamma analysis (2% / 2mm).

Conclusion: Clinical impact: Reduces ERE-driven gamma failures, Reduces false positives, Preserves CCC computation speed, Generalizes across institutions and localizations.