Dortmund 2021 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

ST: Fachverband Strahlen- und Medizinphysik

ST 3: Artificial Intelligence in Medical Physics

ST 3.2: Talk

Tuesday, March 16, 2021, 14:15–14:30, STa

Development of fast dose distribution calculations with generative adversarial networks — Susanna Guatelli², Markus Hagenbuchner², Kevin Kröninger¹, Michael Lerch², •Florian Mentzel¹, Olaf Nackenhorst¹, Jason Paino², Anatoly Rosenfeld², Ayu Saraswati², Ah Chung Tsoi², and Jens Weingarten¹ — ¹TU Dortmund, Germany — ²Centre for Medical Radiation Physics, University of Wollongong, Australia

Radiotherapy targets tumor tissue with radiation to kill cancerous cells. To ensure delivery of the planned dose to the tumor cells while sparing the surrounding healthy tissue, a treatment plan is created before the therapy that defines irradiation angles and durations. The required computation of dose distribution is in many cases accelerated using approximations. For novel irradiation techniques like microbeam radiation therapy (MRT), such approximations have yet to be developed. Therefore, treatment planning is done using time consuming Monte Carlo simulations. A way to create fast dose distribution simulations for novel irradiation modes is the use of generative adversarial networks (GANs). GANs are a class of neural networks that can be trained to generate data points that match the distribution of the training data samples.

We present a study on the development of an algorithm based on 3D-UNet GANs to calculate the dose deriving from minibeam irradiations as simplified case of microbeams irradiation. The dose distributions in different simple target geometries used for the trainings of the GANs were obtained by means of Geant4 simulations.