Weakly supervised 3D ConvLSTMs for high precision Monte-Carlo radiotherapy dose simulations

Sonia Martinot, Norbert Bus, Maria Vakalopoulou, charlotte robert, Eric Deutsch, Nikos Paragios

Abstract: Radiotherapy dose simulation using the Monte-Carlo technique surpasses existing algorithms in terms of precision but remains too time-consuming to be integrated in clinical workflows. We introduce a 3D recurrent and fully convolutional neural network architecture to produce high-precision Monte-Carlo-like dose simulations from low-precision and cheap-to-compute ones. We use the noise-to-noise setting, a weakly supervised training strategy, by training the models solely on low-precision data without expensive-to-compute, high-precision dose simulations. Several evaluation metrics are used to compare with other methods and to assess the clinical viability and quality of the generated dose maps.

Paper Type: both

Primary Subject Area: Application: Other

Secondary Subject Area: Learning with Noisy Labels and Limited Data

Paper Status: original work, not submitted yet

Source Code Url: Our code is under preparation to be publicly available

Data Set Url: Dataset will be made available

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.